Datasets:

sswt

/

arxiv_sample_50K

like 0

Attention UNet as a surrogate model for groundwater prediction ; Numerical simulations of groundwater flow are used to analyze and predict the response of an aquifer system to its change in state by approximating the solution of the fundamental groundwater physical equations. The most used and classical methodologies, such as Finite Difference FD and Finite Element FE Methods, use iterative solvers which are associated with high computational cost. This study proposes a physicsbased convolutional encoderdecoder neural network as a surrogate model to quickly calculate the response of the groundwater system. Holding strong promise in crossdomain mappings, encoderdecoder networks are applicable for learning complex inputoutput mappings of physical systems. This manuscript presents an Attention UNet model that attempts to capture the fundamental inputoutput relations of the groundwater system and generates solutions of hydraulic head in the whole domain given a set of physical parameters and boundary conditions. The model accurately predicts the steady state response of a highly heterogeneous groundwater system given the locations and piezometric head of up to 3 wells as input. The network learns to pay attention only in the relevant parts of the domain and the generated hydraulic head field corresponds to the target samples in great detail. Even relative to coarse finite difference approximations the proposed model is shown to be significantly faster than a comparative stateoftheart numerical solver, thus providing a base for further development of the presented networks as surrogate models for groundwater prediction.

Gaussian dispersion analysis in the time domain efficient conversion with Pade approximants ; We present an approach for adapting the Gaussian dispersion analysis GDA of optical materials to timedomain simulations. Within a GDA model, the imaginary part of a measured dielectric function is presented as a sum of Gaussian absorption terms. Such a simple model is valid for materials where inhomogeneous broadening is substantially larger than the homogeneous linewidth. The GDA model is the essential broadband approximation for the dielectric function of many glasses, polymers, and other natural and artificial materials with disorder. However, efficient implementation of this model in timedomain fullwave electromagnetic solvers has never been fully achieved. We start with a causal form of an isolated oscillator with Gaussiantype absorption Causal DawsonGauss oscillator. Then, we derive explicit analytical formulas to implement the Gaussian oscillator in a finitedifference timedomain FDTD solver with minimal use of memory and floating point operations. The derivation and FDTD implementation employ our generalized dispersive material GDM model a universal, modular approach to describing optical dispersion with Pad'e approximants. We share the FDTD prototype codes that include automated generation of the approximants and a universal FDTD dispersion implementation that employs various secondorder accurate numerical schemes. The codes can be used with noncommercial solvers and commercial software for timedomain simulations of light propagation in dispersive media, which are experimentally characterized with GDA models.

DiffMD A Geometric Diffusion Model for Molecular Dynamics Simulations ; Molecular dynamics MD has long been the de facto choice for simulating complex atomistic systems from first principles. Recently deep learning models become a popular way to accelerate MD. Notwithstanding, existing models depend on intermediate variables such as the potential energy or force fields to update atomic positions, which requires additional computations to perform backpropagation. To waive this requirement, we propose a novel model called DiffMD by directly estimating the gradient of the log density of molecular conformations. DiffMD relies on a scorebased denoising diffusion generative model that perturbs the molecular structure with a conditional noise depending on atomic accelerations and treats conformations at previous timeframes as the prior distribution for sampling. Another challenge of modeling such a conformation generation process is that a molecule is kinetic instead of static, which no prior works have strictly studied. To solve this challenge, we propose an equivariant geometric Transformer as the score function in the diffusion process to calculate corresponding gradients. It incorporates the directions and velocities of atomic motions via 3D spherical FourierBessel representations. With multiple architectural improvements, we outperform stateoftheart baselines on MD17 and isomers of C7O2H10 datasets. This work contributes to accelerating material and drug discovery.

Explainable Fairness in Recommendation ; Existing research on fairnessaware recommendation has mainly focused on the quantification of fairness and the development of fair recommendation models, neither of which studies a more substantial problemidentifying the underlying reason of model disparity in recommendation. This information is critical for recommender system designers to understand the intrinsic recommendation mechanism and provides insights on how to improve model fairness to decision makers. Fortunately, with the rapid development of Explainable AI, we can use model explainability to gain insights into model unfairness. In this paper, we study the problem of explainable fairness, which helps to gain insights about why a system is fair or unfair, and guides the design of fair recommender systems with a more informed and unified methodology. Particularly, we focus on a common setting with featureaware recommendation and exposure unfairness, but the proposed explainable fairness framework is general and can be applied to other recommendation settings and fairness definitions. We propose a Counterfactual Explainable Fairness framework, called CEF, which generates explanations about model fairness that can improve the fairness without significantly hurting the performance.The CEF framework formulates an optimization problem to learn the minimal change of the input features that changes the recommendation results to a certain level of fairness. Based on the counterfactual recommendation result of each feature, we calculate an explainability score in terms of the fairnessutility tradeoff to rank all the featurebased explanations, and select the top ones as fairness explanations.

High precision modeling of polarized signals Moment expansion method generalized to spin2 fields ; The modeling and removal of foregrounds poses a major challenge to searches for signals from inflation using the cosmic microwave background CMB. In particular, the modeling of CMB foregrounds including various spatial averaging effects introduces multiple complications that will have to be accounted for in upcoming analyses. In this work, we introduce the generalization of the intensity moment expansion to the spin2 field of linear polarization the spinmoment expansion. Within this framework, moments become spin2 objects that are directly related to the underlying spectral parameters and polarization angle distribution functions. In obtaining the required expressions for the polarization modeling, we highlight the similarities and differences with the intensity moment methods. A spinor rotation in the complex plane with frequency naturally arises from the first order moment when the signal contains both spectral parameters and polarization angle variations. Additional dependencies are introduced at higher order, and we demonstrate how these can be accounted with several illustrative examples. Our new modeling of the polarized signals reveals to be a powerful tool to model the frequency dependence of the polarization angle. As such, it can be immediately applied to numerous astrophysical situations.

Subverting Fair Image Search with Generative Adversarial Perturbations ; In this work we explore the intersection fairness and robustness in the context of ranking when a ranking model has been calibrated to achieve some definition of fairness, is it possible for an external adversary to make the ranking model behave unfairly without having access to the model or training data To investigate this question, we present a case study in which we develop and then attack a stateoftheart, fairnessaware image search engine using images that have been maliciously modified using a Generative Adversarial Perturbation GAP model. These perturbations attempt to cause the fair reranking algorithm to unfairly boost the rank of images containing people from an adversaryselected subpopulation. We present results from extensive experiments demonstrating that our attacks can successfully confer significant unfair advantage to people from the majority class relative to fairlyranked baseline search results. We demonstrate that our attacks are robust across a number of variables, that they have close to zero impact on the relevance of search results, and that they succeed under a strict threat model. Our findings highlight the danger of deploying fair machine learning algorithms inthewild when 1 the data necessary to achieve fairness may be adversarially manipulated, and 2 the models themselves are not robust against attacks.

On learning agentbased models from data ; AgentBased Models ABMs are used in several fields to study the evolution of complex systems from microlevel assumptions. However, ABMs typically can not estimate agentspecific or micro variables this is a major limitation which prevents ABMs from harnessing microlevel data availability and which greatly limits their predictive power. In this paper, we propose a protocol to learn the latent microvariables of an ABM from data. The first step of our protocol is to reduce an ABM to a probabilistic model, characterized by a computationally tractable likelihood. This reduction follows two general design principles balance of stochasticity and data availability, and replacement of unobservable discrete choices with differentiable approximations. Then, our protocol proceeds by maximizing the likelihood of the latent variables via a gradientbased expectation maximization algorithm. We demonstrate our protocol by applying it to an ABM of the housing market, in which agents with different incomes bid higher prices to live in highincome neighborhoods. We demonstrate that the obtained model allows accurate estimates of the latent variables, while preserving the general behavior of the ABM. We also show that our estimates can be used for outofsample forecasting. Our protocol can be seen as an alternative to blackbox data assimilation methods, that forces the modeler to lay bare the assumptions of the model, to think about the inferential process, and to spot potential identification problems.

Modelling spatially autocorrelated detection probabilities in spatial capturerecapture using random effects ; Spatial capturerecapture SCR models are now widely used for estimating density from repeated individual spatial encounters. SCR accounts for the inherent spatial autocorrelation in individual detections by modelling detection probabilities as a function of distance between the detectors and individual activity centres. However, additional spatial heterogeneity in detection probability may still creep in due to environmental or sampling characteristics. if unaccounted for, such variation can lead to pronounced bias in population size estimates. Using simulations, we describe and test three Bayesian SCR models that use generalized linear mixed models GLMM to account for latent heterogeneity in baseline detection probability across detectors using independent random effects RE, spatially autocorrelated random effects SARE, and a twogroup finite mixture model FM. Overall, SARE provided the least biased population size estimates median RB 9 6. When spatial autocorrelation was high, SARE also performed best at predicting the spatial pattern of heterogeneity in detection probability. At intermediate levels of autocorrelation, spatiallyexplicit estimates of detection probability obtained with FM where more accurate than those generated by SARE and RE. In cases where the number of detections per detector is realistically low at most 1, all GLMMs considered here may require dimension reduction of the random effects by pooling baseline detection probability parameters across neighboring detectors aggregation to avoid overparameterization. The added complexity and computational overhead associated with SCRGLMMs may only be justified in extreme cases of spatial heterogeneity. However, even in less extreme cases, detecting and estimating spatially heterogeneous detection probability may assist in planning or adjusting monitoring schemes.

Modeling Human Behavior Part II Cognitive approaches and Uncertainty ; As we discussed in Part I of this topic, there is a clear desire to model and comprehend human behavior. Given the popular presupposition of human reasoning as the standard for learning and decisionmaking, there have been significant efforts and a growing trend in research to replicate these innate human abilities in artificial systems. In Part I, we discussed learning methods which generate a model of behavior from exploration of the system and feedback based on the exhibited behavior as well as topics relating to the use of or accounting for beliefs with respect to applicable skills or mental states of others. In this work, we will continue the discussion from the perspective of methods which focus on the assumed cognitive abilities, limitations, and biases demonstrated in human reasoning. We will arrange these topics as follows i methods such as cognitive architectures, cognitive heuristics, and related which demonstrate assumptions of limitations on cognitive resources and how that impacts decisions and ii methods which generate and utilize representations of bias or uncertainty to model human decisionmaking or the future outcomes of decisions.

GenerSpeech Towards Style Transfer for Generalizable OutOfDomain TexttoSpeech ; Style transfer for outofdomain OOD speech synthesis aims to generate speech samples with unseen style e.g., speaker identity, emotion, and prosody derived from an acoustic reference, while facing the following challenges 1 The highly dynamic style features in expressive voice are difficult to model and transfer; and 2 the TTS models should be robust enough to handle diverse OOD conditions that differ from the source data. This paper proposes GenerSpeech, a texttospeech model towards highfidelity zeroshot style transfer of OOD custom voice. GenerSpeech decomposes the speech variation into the styleagnostic and stylespecific parts by introducing two components 1 a multilevel style adaptor to efficiently model a large range of style conditions, including global speaker and emotion characteristics, and the local utterance, phoneme, and wordlevel finegrained prosodic representations; and 2 a generalizable content adaptor with MixStyle Layer Normalization to eliminate style information in the linguistic content representation and thus improve model generalization. Our evaluations on zeroshot style transfer demonstrate that GenerSpeech surpasses the stateoftheart models in terms of audio quality and style similarity. The extension studies to adaptive style transfer further show that GenerSpeech performs robustly in the fewshot data setting. Audio samples are available at httpsGenerSpeech.github.io

Chirality of Gravitational Waves in ChernSimons fR Gravity Cosmology ; In this paper we shall consider an axionic ChernSimons corrected fR gravity theoretical framework, and we shall study the chirality of the generated primordial gravitational waves. Particularly, we shall consider two main axion models, the canonical misalignment axion model and the kinetic axion model, both of which provide an interesting particle phenomenology, in the presence of R2 terms in the inflationary Lagrangian. The axion does not affect significantly the background evolution during the inflationary era, which is solely controlled by R2 gravity. However, the due to the presence of the ChernSimons term, the tensor perturbations are directly affected, and our aim is to reveal the extent of effects of the ChernSimons term on the gravitational waves modes, for both the axion models. We aim to produce analytical descriptions of the primordial tensor modes behavior, and thus we solve analytically the evolution equations of the tensor modes, for a nearly de Sitter primordial evolution controlled by the R2 gravity. We focus the analytical study on superhorizon and subhorizon modes. For the misalignment model, we were able to produce analytic solutions for both the subhorizon and superhorizon modes, in which case we found the behavior of the circular polarization function. Our results indicate that the produced tensor spectrum is strongly chiral. For the kinetic axion model though, analytic solutions are obtained only for the superhorizon modes. In order to have a grasp of the behavior of the chirality of the tensor modes, we studied the chirality of the superhorizon modes, however a more complete study is needed, which is impossible to do analytically though.

Functional Ensemble Distillation ; Bayesian models have many desirable properties, most notable is their ability to generalize from limited data and to properly estimate the uncertainty in their predictions. However, these benefits come at a steep computational cost as Bayesian inference, in most cases, is computationally intractable. One popular approach to alleviate this problem is using a MonteCarlo estimation with an ensemble of models sampled from the posterior. However, this approach still comes at a significant computational cost, as one needs to store and run multiple models at test time. In this work, we investigate how to best distill an ensemble's predictions using an efficient model. First, we argue that current approaches that simply return distribution over predictions cannot compute important properties, such as the covariance between predictions, which can be valuable for further processing. Second, in many limited data settings, all ensemble members achieve nearly zero training loss, namely, they produce nearidentical predictions on the training set which results in suboptimal distilled models. To address both problems, we propose a novel and general distillation approach, named Functional Ensemble Distillation FED, and we investigate how to best distill an ensemble in this setting. We find that learning the distilled model via a simple augmentation scheme in the form of mixup augmentation significantly boosts the performance. We evaluated our method on several tasks and showed that it achieves superior results in both accuracy and uncertainty estimation compared to current approaches.

Making Large Language Models Better Reasoners with StepAware Verifier ; Fewshot learning is a challenging task that requires language models to generalize from limited examples. Large language models like GPT3 and PaLM have made impressive progress in this area, but they still face difficulties in reasoning tasks such as GSM8K, a benchmark for arithmetic problems. To improve their reasoning skills, previous work has proposed to guide the language model with prompts that elicit a series of reasoning steps before giving the final answer, achieving a significant improvement on GSM8K from 17.9 to 58.1 in problemsolving rate. In this paper, we present DIVERSE Diverse Verifier on Reasoning Step, a novel approach that further enhances the reasoning capability of language models. DIVERSE has three main components first, it generates diverse prompts to explore different reasoning paths for the same question; second, it uses a verifier to filter out incorrect answers based on a weighted voting scheme; and third, it verifies each reasoning step individually instead of the whole chain. We evaluate DIVERSE on the latest language model codedavinci002 and show that it achieves new stateoftheart results on six of eight reasoning benchmarks e.g., GSM8K 74.4 to 83.2.

PhysicsInspired Temporal Learning of Quadrotor Dynamics for Accurate Model Predictive Trajectory Tracking ; Accurately modeling quadrotor's system dynamics is critical for guaranteeing agile, safe, and stable navigation. The model needs to capture the system behavior in multiple flight regimes and operating conditions, including those producing highly nonlinear effects such as aerodynamic forces and torques, rotor interactions, or possible system configuration modifications. Classical approaches rely on handcrafted models and struggle to generalize and scale to capture these effects. In this paper, we present a novel PhysicsInspired Temporal Convolutional Network PITCN approach to learning quadrotor's system dynamics purely from robot experience. Our approach combines the expressive power of sparse temporal convolutions and dense feedforward connections to make accurate system predictions. In addition, physics constraints are embedded in the training process to facilitate the network's generalization capabilities to data outside the training distribution. Finally, we design a model predictive control approach that incorporates the learned dynamics for accurate closedloop trajectory tracking fully exploiting the learned model predictions in a receding horizon fashion. Experimental results demonstrate that our approach accurately extracts the structure of the quadrotor's dynamics from data, capturing effects that would remain hidden to classical approaches. To the best of our knowledge, this is the first time physicsinspired deep learning is successfully applied to temporal convolutional networks and to the system identification task, while concurrently enabling predictive control.

Predictions of Electromotive Force of Magnetic Shape Memory Alloy MSMA Using Constitutive Model and Generalized Regression Neural Network ; Ferromagnetic shape memory alloys MSMAs, such as NiMnGa single crystals, can exhibit the shape memory effect due to an applied magnetic field at room temperature. Under a variable magnetic field and a constant bias stress loading, MSMAs have been used for actuation applications. This work introduced a new feature to the existing macroscale magnetomechanical model for NiMnGa single crystal. This model includes the fact that the magnetic easy axis in the two variants is not exactly perpendicular as observed by D silva et al. This offset helps explain some of the power harvesting capabilities of MSMAs. Model predictions are compared to experimental data collected on a NiMnGa single crystal. The experiments include both stresscontrolled loading with constant bias magnetic field load which mimics power harvesting or sensing and fieldcontrolled loading with constant bias compressive stress which mimics actuation. Each type of test was performed at several different load levels, and the applied field was measured without the MSMA specimen present so that demagnetization does not affect the experimentally measured field as suggested by Eberle et al. Results show decent agreement between model predictions and experimental data. Although the model predicts experimental results decently, it does not capture all the features of the experimental data. In order to capture all the experimental features, finally, a generalized regression neural network GRNN was used to train the experimental data stress, strain, magnetic field, and emf so that it can make a reasonably better prediction.

Multifidelity Hierarchical Neural Processes ; Science and engineering fields use computer simulation extensively. These simulations are often run at multiple levels of sophistication to balance accuracy and efficiency. Multifidelity surrogate modeling reduces the computational cost by fusing different simulation outputs. Cheap data generated from lowfidelity simulators can be combined with limited highquality data generated by an expensive highfidelity simulator. Existing methods based on Gaussian processes rely on strong assumptions of the kernel functions and can hardly scale to highdimensional settings. We propose Multifidelity Hierarchical Neural Processes MFHNP, a unified neural latent variable model for multifidelity surrogate modeling. MFHNP inherits the flexibility and scalability of Neural Processes. The latent variables transform the correlations among different fidelity levels from observations to latent space. The predictions across fidelities are conditionally independent given the latent states. It helps alleviate the error propagation issue in existing methods. MFHNP is flexible enough to handle nonnested high dimensional data at different fidelity levels with varying input and output dimensions. We evaluate MFHNP on epidemiology and climate modeling tasks, achieving competitive performance in terms of accuracy and uncertainty estimation. In contrast to deep Gaussian Processes with only lowdimensional 10 tasks, our method shows great promise for speeding up highdimensional complex simulations over 7000 for epidemiology modeling and 45000 for climate modeling.

Deep Leakage from Model in Federated Learning ; Distributed machine learning has been widely used in recent years to tackle the large and complex dataset problem. Therewith, the security of distributed learning has also drawn increasing attentions from both academia and industry. In this context, federated learning FL was developed as a secure distributed learning by maintaining private training data locally and only public model gradients are communicated between. However, to date, a variety of gradient leakage attacks have been proposed for this procedure and prove that it is insecure. For instance, a common drawback of these attacks is shared they require too much auxiliary information such as model weights, optimizers, and some hyperparameters e.g., learning rate, which are difficult to obtain in real situations. Moreover, many existing algorithms avoid transmitting model gradients in FL and turn to sending model weights, such as FedAvg, but few people consider its security breach. In this paper, we present two novel frameworks to demonstrate that transmitting model weights is also likely to leak private local data of clients, i.e., DLM and DLM, under the FL scenario. In addition, a number of experiments are performed to illustrate the effect and generality of our attack frameworks. At the end of this paper, we also introduce two defenses to the proposed attacks and evaluate their protection effects. Comprehensively, the proposed attack and defense schemes can be applied to the general distributed learning scenario as well, just with some appropriate customization.

How Much is Enough A Study on Diffusion Times in Scorebased Generative Models ; Scorebased diffusion models are a class of generative models whose dynamics is described by stochastic differential equations that map noise into data. While recent works have started to lay down a theoretical foundation for these models, an analytical understanding of the role of the diffusion time T is still lacking. Current best practice advocates for a large T to ensure that the forward dynamics brings the diffusion sufficiently close to a known and simple noise distribution; however, a smaller value of T should be preferred for a better approximation of the scorematching objective and higher computational efficiency. Starting from a variational interpretation of diffusion models, in this work we quantify this tradeoff, and suggest a new method to improve quality and efficiency of both training and sampling, by adopting smaller diffusion times. Indeed, we show how an auxiliary model can be used to bridge the gap between the ideal and the simulated forward dynamics, followed by a standard reverse diffusion process. Empirical results support our analysis; for image data, our method is competitive w.r.t. the stateoftheart, according to standard sample quality metrics and loglikelihood.

Fast building segmentation from satellite imagery and few local labels ; Innovations in computer vision algorithms for satellite image analysis can enable us to explore global challenges such as urbanization and land use change at the planetary level. However, domain shift problems are a common occurrence when trying to replicate models that drive these analyses to new areas, particularly in the developing world. If a model is trained with imagery and labels from one location, then it usually will not generalize well to new locations where the content of the imagery and data distributions are different. In this work, we consider the setting in which we have a single large satellite imagery scene over which we want to solve an applied problem building footprint segmentation. Here, we do not necessarily need to worry about creating a model that generalizes past the borders of our scene but can instead train a local model. We show that surprisingly few labels are needed to solve the building segmentation problem with very highresolution 0.5mpx satellite imagery with this setting in mind. Our best model trained with just 527 sparse polygon annotations an equivalent of 1500 x 1500 densely labeled pixels has a recall of 0.87 over held out footprints and a R2 of 0.93 on the task of counting the number of buildings in 200 x 200meter windows. We apply our models over highresolution imagery in Amman, Jordan in a case study on urban change detection.

The Response of Dark Matter Haloes to Gas Ejection CuspCore II ; We propose an analytic model, CuspCore II, for the response of dark matter DM haloes to central gas ejection, as a mechanism for generating DMdeficient cores in dwarfs and highz massive galaxies. We test this model and three other methods using idealized Nbody simulations. The current model is physically justified and provides more accurate predictions than the earlier version, CuspCore I Freundlich et al. 2020. The CuspCore model assumes an instantaneous change of potential, followed by a relaxation to a new Jeans equilibrium. The relaxation turns out to be violent relaxation during the first orbital period, followed by phase mixing. By tracing the energy diffusion dEdUr iteratively, the model reproduces the simulated DM profiles with 10 accuracy or better. A method based on adiabatic invariants shows similar precision for moderate mass change but underestimates the DM expansion for strong gas ejection. A method based on a simple empirical relation between DM and total mass ratios makes slightly inferior predictions. The crude assumption used in CuspCore I, of energy conservation for shells that encompass a fixed DM mass, turns out to underestimate the DM response, which can be partially remedied by introducing an alternative energy definition. Our model is being generalized to address the differential response of a multicomponent system of stars and DM in the formation of DMdeficient galaxies.

Investigation of g2 anomaly in the specific 2HDM with Vector like leptons and the phenomenological implications ; The anomalous magnetic moment of muons has been a longstanding problem in SM. The current deviation of experimental value of the g2mu from the standard model prediction is exactly 4.2sigma. Two Higgs Doublet Models can accommodate this discrepancy but such type of model naturally generate flavor changing neutral currentFCNC. To prevent this it was postulated that 2HDM without FCNC required that all fermions of a given charge couple to the same Higgs boson but the rule breaks in Muon Specific Two Higgs Doublet Model where all fermions except muon couple to one Higgs doublet and muon with the other Higgs doublet. The Muon Specific Two Higgs Doublet model explain muon anomaly with a fine tuning problem of very large tanbeta value with other parameters. We have found a simple solution of this fine tuning problem by extending this model with a vector like lepton generation which could explain the muon anomaly at low tanbeta value with a heavy pseudo scalar Higgs boson under the shadow of current experimental and theoretical constraints. Moreover, with the help of the cut based analysis and multivariate analysis methods, we have also attempted to shed some light on the potential experimental signature of vector lepton decay to the heavy Higgs boson in the LHC experiment. We have showed that a multivariate analysis can increase the vector like leptons signal significance by up to an order of magnitude than that of a cut based analysis.

Classification and Generation of realworld data with an Associative Memory Model ; Drawing from memory the face of a friend you have not seen in years is a difficult task. However, if you happen to cross paths, you would easily recognize each other. The biological memory is equipped with an impressive compression algorithm that can store the essential, and then infer the details to match perception. The Willshaw Memory is a simple abstract model for cortical computations which implements mechanisms of biological memories. Using our recently proposed sparse coding prescription for visual patterns, this model can store and retrieve an impressive amount of realworld data in a faulttolerant manner. In this paper, we extend the capabilities of the basic Associative Memory Model by using a MultipleModality framework. In this setting, the memory stores several modalities e.g., visual, or textual of each pattern simultaneously. After training, the memory can be used to infer missing modalities when just a subset is perceived. Using a simple encodermemorydecoder architecture, and a newly proposed iterative retrieval algorithm for the Willshaw Model, we perform experiments on the MNIST dataset. By storing both the images and labels as modalities, a single Memory can be used not only to retrieve and complete patterns but also to classify and generate new ones. We further discuss how this model could be used for other learning tasks, thus serving as a biologicallyinspired framework for learning.

Scaleinvariant 3311 model with BL symmetry ; Motivated by a possible interplay between the mechanism of dynamical symmetry breaking and the seesaw mechanism for generating fermion masses, we present a scaleinvariant model that extends the gauge symmetry of the Standard Model electroweak sector to SU3LotimesU1XotimesU1N, with a builtin BL symmetry. The model is based on the symmetry structure of the known 331 models and, thus, it relates the number of the three observed fermion generations with the cancellation of gauge anomalies. Symmetry breaking is triggered via the ColemanWeinberg mechanism taking into account a minimal set of scalar field multiplets. We establish the stability conditions for the treelevel scalar potential imposing the copositivity criteria and use the method of GildenerWeinberg for computing the oneloop effective potential when one has multiple scalar fields. With the addition of vectorial fermions, getting their mass mainly through the vacuum expectation value of scalar singlets at 103 TeV, the BL symmetry leads to textures for the fermion mass matrices, allowing seesaw mechanisms for neutrinos and quarks to take place. In particular, these mechanisms could partly explain the mass hierarchies of the quarks. Once the breakdown of the SU3L symmetry is supposed to occur around 10 TeV, the model also predicts new particles with TeVscale masses, such as a neutral scalar, H1, a charged scalar, Hpm, and the gauge bosons Zprime, Wprimepm and Y0, that could be searched with the highluminosity LHC.

InterferenceLimited UltraReliable and LowLatency Communications Graph Neural Networks or Stochastic Geometry ; In this paper, we aim to improve the QualityofService QoS of UltraReliability and LowLatency Communications URLLC in interferencelimited wireless networks. To obtain time diversity within the channel coherence time, we first put forward a random repetition scheme that randomizes the interference power. Then, we optimize the number of reserved slots and the number of repetitions for each packet to minimize the QoS violation probability, defined as the percentage of users that cannot achieve URLLC. We build a cascaded Random Edge Graph Neural Network REGNN to represent the repetition scheme and develop a modelfree unsupervised learning method to train it. We analyze the QoS violation probability using stochastic geometry in a symmetric scenario and apply a modelbased Exhaustive Search ES method to find the optimal solution. Simulation results show that in the symmetric scenario, the QoS violation probabilities achieved by the modelfree learning method and the modelbased ES method are nearly the same. In more general scenarios, the cascaded REGNN generalizes very well in wireless networks with different scales, network topologies, cell densities, and frequency reuse factors. It outperforms the modelbased ES method in the presence of the model mismatch.

Spatial point process via regularisation modelling of ambulance call risk ; This study investigates the spatial distribution of emergency alarm call events to identify spatial covariates associated with the events and discern hotspot regions for the events. The study is motivated by the problem of developing optimal dispatching strategies for prehospital resources such as ambulances. To achieve our goals, we model the spatially varying call occurrence risk as an intensity function of an inhomogeneous spatial Poisson process that we assume is a loglinear function of some underlying spatial covariates. The spatial covariates used in this study are related to road network coverage, population density, and the socioeconomic status of the population in Skellefteaa, Sweden. A new heuristic algorithm has been developed to select an optimal estimate of the kernel bandwidth in order to obtain the nonparametric intensity estimate of the events and to generate other covariates. Since we consider a large number of spatial covariates as well as their products, and since some of them may be strongly correlated, lassolike elasticnet regularisation has been used in the loglikelihood intensity modeling to perform variable selection and reduce variance inflation from overfitting and bias from underfitting. As a result of the variable selection, the fitted model structure contains individual covariates of both road network and demographic types. We discovered that hotspot regions of calls have been observed along dense parts of the road network. Evaluation of the model also suggests that the estimated model is stable and can be used to generate a reliable intensity estimate over the region, which can be used as an input in the problem of designing prehospital resource dispatching strategies.

Multiple Robust Learning for Recommendation ; In recommender systems, a common problem is the presence of various biases in the collected data, which deteriorates the generalization ability of the recommendation models and leads to inaccurate predictions. Doubly robust DR learning has been studied in many tasks in RS, with the advantage that unbiased learning can be achieved when either a single imputation or a single propensity model is accurate. In this paper, we propose a multiple robust MR estimator that can take the advantage of multiple candidate imputation and propensity models to achieve unbiasedness. Specifically, the MR estimator is unbiased when any of the imputation or propensity models, or a linear combination of these models is accurate. Theoretical analysis shows that the proposed MR is an enhanced version of DR when only having a single imputation and propensity model, and has a smaller bias. Inspired by the generalization error bound of MR, we further propose a novel multiple robust learning approach with stabilization. We conduct extensive experiments on realworld and semisynthetic datasets, which demonstrates the superiority of the proposed approach over stateoftheart methods.

Enhancing Image Rescaling using Dual Latent Variables in Invertible Neural Network ; Normalizing flow models have been used successfully for generative image superresolution SR by approximating complex distribution of natural images to simple tractable distribution in latent space through Invertible Neural Networks INN. These models can generate multiple realistic SR images from one lowresolution LR input using randomly sampled points in the latent space, simulating the illposed nature of image upscaling where multiple highresolution HR images correspond to the same LR. Lately, the invertible process in INN has also been used successfully by bidirectional image rescaling models like IRN and HCFlow for joint optimization of downscaling and inverse upscaling, resulting in significant improvements in upscaled image quality. While they are optimized for image downscaling too, the illposed nature of image downscaling, where one HR image could be downsized to multiple LR images depending on different interpolation kernels and resampling methods, is not considered. A new downscaling latent variable, in addition to the original one representing uncertainties in image upscaling, is introduced to model variations in the image downscaling process. This dual latent variable enhancement is applicable to different image rescaling models and it is shown in extensive experiments that it can improve image upscaling accuracy consistently without sacrificing image quality in downscaled LR images. It is also shown to be effective in enhancing other INNbased models for image restoration applications like image hiding.

REPNP PlugandPlay with Deep Reinforcement Learning Prior for Robust Image Restoration ; Image restoration schemes based on the pretrained deep models have received great attention due to their unique flexibility for solving various inverse problems. In particular, the PlugandPlay PnP framework is a popular and powerful tool that can integrate an offtheshelf deep denoiser for different image restoration tasks with known observation models. However, obtaining the observation model that exactly matches the actual one can be challenging in practice. Thus, the PnP schemes with conventional deep denoisers may fail to generate satisfying results in some realworld image restoration tasks. We argue that the robustness of the PnP framework is largely limited by using the offtheshelf deep denoisers that are trained by deterministic optimization. To this end, we propose a novel deep reinforcement learning DRL based PnP framework, dubbed RePNP, by leveraging a lightweight DRLbased denoiser for robust image restoration tasks. Experimental results demonstrate that the proposed RePNP is robust to the observation model used in the PnP scheme deviating from the actual one. Thus, RePNP can generate more reliable restoration results for image deblurring and super resolution tasks. Compared with several stateoftheart deep image restoration baselines, RePNP achieves better results subjective to model deviation with fewer model parameters.

Staggered grids for multidimensional multiscale modelling ; Numerical schemes for wavelike systems with small dissipation are often inaccurate and unstable due to truncation errors and numerical roundoff errors. Hence, numerical simulations of wavelike systems lacking proper handling of these numerical issues often fail to represent the physical characteristics of wave phenomena. This challenge gets even more intricate for multiscale modelling, especially in multiple dimensions. When using the usual collocated grid, about twothirds of the resolved wave modes are incorrect with significant dispersion. But, numerical schemes on staggered grids with alternating variable arrangement are significantly less dispersive and preserve much of the wave characteristics. Also, the group velocity of the energy propagation in the numerical waves on a staggered grid is in the correct direction, in contrast to the collocated grid. For high accuracy and to preserve much of the wave characteristics, this article extends the concept of staggered grids in fulldomain modelling to multidimensional multiscale modelling. Specifically, this article develops 120 multiscale staggered grids and demonstrates their stability, accuracy, and wavepreserving characteristic for equationfree multiscale modelling of weakly damped linear waves. But most characteristics of the developed multiscale staggered grids must also hold in general for multiscale modelling of many complex spatiotemporal physical phenomena such as the general computational fluid dynamics.

Neural Knowledge Bank for Pretrained Transformers ; The ability of pretrained Transformers to remember factual knowledge is essential but still limited for existing models. Inspired by existing work that regards FeedForward Networks FFNs in Transformers as keyvalue memories, we design a Neural Knowledge Bank NKB and a knowledge injection strategy to introduce extra factual knowledge for pretrained Transformers. The NKB is in the form of additional knowledgeable memory slots to the FFN and the memorylike architecture makes it highly interpretable and flexible. When injecting extra knowledge with the Salient Span Masking SSM pretraining objective, we fix the original pretrained model and train only the NKB. This training strategy makes sure the general language modeling ability of the original pretrained model is not influenced. By mounting the NKB onto the T5 model, we verify its strong ability to store extra factual knowledge based on three closedbook question answering datasets. Also, we prove that mounting the NKB will not degrade the general language modeling ability of T5 through two representative tasks, summarization and machine translation. Further, we thoroughly analyze the interpretability of the NKB and reveal the meaning of its keys and values in a humanreadable way. Finally, we show the flexibility of the NKB by directly modifying its value vectors to update the factual knowledge stored in it.

Planning ridepooling services with detour restrictions for spatially heterogeneous demand A multizone queuing network approach ; This study presents a multizone queuing network model for steadystate ridepooling operations that serve heterogeneous demand, and then builds upon this model to optimize the design of ridepooling services. Spatial heterogeneity is addressed by partitioning the study region into a set of relatively homogeneous zones, and a set of criteria are imposed to avoid significant detours among matched passengers. A generalized multizone queuing network model is then developed to describe how vehicles' states transition within each zone and across neighboring zones, and how passengers are served by idle or partially occupied vehicles. A large system of equations is constructed based on the queuing network model to analytically evaluate steadystate system performance. Then, we formulate a constrained nonlinear program to optimize the design of ridepooling services, such as zonelevel vehicle deployment, vehicle routing paths, and vehicle rebalancing operations. A customized solution approach is also proposed to decompose and solve the optimization problem. The proposed model and solution approach are applied to a hypothetical case and a realworld Chicago case study, so as to demonstrate their applicability and to draw insights. Agentbased simulations are also used to corroborate results from the proposed analytical model. These numerical examples not only reveal interesting insights on how ridepooling services serve heterogeneous demand, but also highlight the importance of addressing demand heterogeneity when designing ridepooling services.

Federated Adversarial Learning A Framework with Convergence Analysis ; Federated learning FL is a trending training paradigm to utilize decentralized training data. FL allows clients to update model parameters locally for several epochs, then share them to a global model for aggregation. This training paradigm with multilocal step updating before aggregation exposes unique vulnerabilities to adversarial attacks. Adversarial training is a popular and effective method to improve the robustness of networks against adversaries. In this work, we formulate a general form of federated adversarial learning FAL that is adapted from adversarial learning in the centralized setting. On the client side of FL training, FAL has an inner loop to generate adversarial samples for adversarial training and an outer loop to update local model parameters. On the server side, FAL aggregates local model updates and broadcast the aggregated model. We design a global robust training loss and formulate FAL training as a minmax optimization problem. Unlike the convergence analysis in classical centralized training that relies on the gradient direction, it is significantly harder to analyze the convergence in FAL for three reasons 1 the complexity of minmax optimization, 2 model not updating in the gradient direction due to the multilocal updates on the clientside before aggregation and 3 interclient heterogeneity. We address these challenges by using appropriate gradient approximation and coupling techniques and present the convergence analysis in the overparameterized regime. Our main result theoretically shows that the minimum loss under our algorithm can converge to epsilon small with chosen learning rate and communication rounds. It is noteworthy that our analysis is feasible for nonIID clients.

Deep Learning Driven Natural Languages Text to SQL Query Conversion A Survey ; With the future striving toward datacentric decisionmaking, seamless access to databases is of utmost importance. There is extensive research on creating an efficient texttosql TEXT2SQL model to access data from the database. Using a Natural language is one of the best interfaces that can bridge the gap between the data and results by accessing the database efficiently, especially for nontechnical users. It will open the doors and create tremendous interest among users who are well versed in technical skills or not very skilled in query languages. Even if numerous deep learningbased algorithms are proposed or studied, there still is very challenging to have a generic model to solve the data query issues using natural language in a realwork scenario. The reason is the use of different datasets in different studies, which comes with its limitations and assumptions. At the same time, we do lack a thorough understanding of these proposed models and their limitations with the specific dataset it is trained on. In this paper, we try to present a holistic overview of 24 recent neural network models studied in the last couple of years, including their architectures involving convolutional neural networks, recurrent neural networks, pointer networks, reinforcement learning, generative models, etc. We also give an overview of the 11 datasets that are widely used to train the models for TEXT2SQL technologies. We also discuss the future application possibilities of TEXT2SQL technologies for seamless data queries.

RuDi Explaining Behavior Sequence Models by Automatic Statistics Generation and Rule Distillation ; Risk scoring systems have been widely deployed in many applications, which assign risk scores to users according to their behavior sequences. Though many deep learning methods with sophisticated designs have achieved promising results, the blackbox nature hinders their applications due to fairness, explainability, and compliance consideration. Rulebased systems are considered reliable in these sensitive scenarios. However, building a rule system is laborintensive. Experts need to find informative statistics from user behavior sequences, design rules based on statistics and assign weights to each rule. In this paper, we bridge the gap between effective but blackbox models and transparent rule models. We propose a twostage method, RuDi, that distills the knowledge of blackbox teacher models into rulebased student models. We design a Monte Carlo tree searchbased statistics generation method that can provide a set of informative statistics in the first stage. Then statistics are composed into logical rules with our proposed neural logical networks by mimicking the outputs of teacher models. We evaluate RuDi on three realworld public datasets and an industrial dataset to demonstrate its effectiveness.

Datadriven modeling of beam loss in the LHC ; In the Large Hadron Collider, the beam losses are continuously measured for machine protection. By design, most of the particle losses occur in the collimation system, where the particles with high oscillation amplitudes or large momentum error are scraped from the beams. The level of particle losses typically is optimized manually by changing multiple control parameters, among which are, for example, currents in the focusing and defocusing magnets along the collider. It is generally challenging to model and predict losses based on the control parameters due to various nonlinear effects in the system, such as electron clouds, resonance effects, etc, and multiple sources of uncertainty. At the same time understanding the influence of control parameters on the losses is extremely important in order to improve the operation and performance, and future design of accelerators. Existing results showed that it is hard to generalize the models, which assume the regression model of losses depending on control parameters, from fills carried out throughout one year to the data of another year. To circumvent this, we propose to use an autoregressive modeling approach, where we take into account not only the observed control parameters but also previous loss values. We use an equivalent Kalman Filter KF formulation in order to efficiently estimate models with different lags.

Dyonic Matter Equations, Exact PointSource Solutions, and Charged Black Holes in Generalized BornInfeld Theory ; We derive the equations of motion governing static dyonic matters, described in terms of two real scalar fields, in nonlinear electrodynamics of the BornInfeld theory type. We then obtain exact finiteenergy solutions of these equations in the quadratic and logarithmic nonlinearity cases subject to dyonic pointcharge sources and construct dyonically charged black holes with relegated curvature singularities. In the case of quadratic nonlinearity, which is the core model of this work, we show that dyonic solutions enable us to restore electromagnetic symmetry, which is known to be broken in nondyonic situations by exclusion of monopoles. We further demonstrate that in the context of kessence cosmology the nonlinear electrodynamics models possess their own distinctive signatures in light of the underlying equations of state of the cosmic fluids they represent. In this context, the quadratic and logarithmic models are shown to resolve a densitypressure inconsistency issue exhibited by the original BornInfeld model kessence action function as well as by all of its fractionalpowered extensions. Moreover, it is shown that the quadratic model is uniquely positioned to give rise to a radiationdominated era in the early universe among all the polynomial models and other examples considered.

Attractor Stability in Finite Asynchronous Biological System Models ; We present mathematical techniques for exhaustive studies of longterm dynamics of asynchronous biological system models. Specifically, we extend the notion of kappaequivalence developed for graph dynamical systems to support systematic analysis of all possible attractor configurations that can be generated when varying the asynchronous update order Macauley and Mortveit 2009. We extend earlier work by VelizCuba and Stigler 2011, Goles et al. 2014, and others by comparing longterm dynamics up to topological conjugation rather than comparing the exact states and their transitions on attractors, we only compare the attractor structures. In general, obtaining this information is computationally intractable. Here, we adapt and apply combinatorial theory for dynamical systems to develop computational methods that greatly reduce this computational cost. We give a detailed algorithm and apply it to i the lac operon model for Escherichia coli proposed by VelizCuba and Stigler 2011, and ii the regulatory network involved in the control of the cell cycle and cell differentiation in the Caenorhabditis elegans vulva precursor cells proposed by Weinstein et al. 2015. In both cases, we uncover all possible limit cycle structures for these networks under sequential updates. Specifically, for the lac operon model, rather than examining all 10 3.6 cdot 106 sequential update orders, we demonstrate that it is sufficient to consider 344 representative update orders, and, more notably, that these 344 representatives give rise to 4 distinct attractor structures. A similar analysis performed for the C. elegans model demonstrates that it has precisely 125 distinct attractor structures. We conclude with observations on the variety and distribution of the models' attractor structures and use the results to discuss their robustness.

Leveraging LowFidelity Data to Improve Machine Learning of Sparse HighFidelity Thermal Conductivity Data via Transfer Learning ; Lattice thermal conductivity TC of semiconductors is crucial for various applications, ranging from microelectronics to thermoelectrics. Datadriven approach can potentially establish the critical compositionproperty relationship needed for fast screening of candidates with desirable TC, but the small number of available data remains the main challenge. TC can be efficiently calculated using empirical models, but they have inferior accuracy compared to the more resourcedemanding firstprinciples calculations. Here, we demonstrate the use of transfer learning TL to improve the machine learning models trained on small but highfidelity TC data from experiments and firstprinciples calculations, by leveraging a large but lowfidelity data generated from empirical TC models, where the trainings on high and lowfidelity TC data are treated as different but related tasks. TL improves the model accuracy by as much as 23 in R2 and reduces the average factor difference by as much as 30. Using the TL model, a large semiconductor database is screened, and several candidates with room temperature TC 350 WmK are identified and further verified using firstprinciples simulations. This study demonstrates that TL can leverage big lowfidelity data as a proxy task to improve models for the target task with highfidelity but small data. Such a capability of TL may have important implications to materials informatics in general.

Predicting IMDb Rating of TV Series with Deep Learning The Case of Arrow ; Context The number of TV series offered nowadays is very high. Due to its large amount, many series are canceled due to a lack of originality that generates a low audience. Problem Having a decision support system that can show why some shows are a huge success or not would facilitate the choices of renewing or starting a show. Solution We studied the case of the series Arrow broadcasted by CW Network and used descriptive and predictive modeling techniques to predict the IMDb rating. We assumed that the theme of the episode would affect its evaluation by users, so the dataset is composed only by the director of the episode, the number of reviews that episode got, the percentual of each theme extracted by the Latent Dirichlet Allocation LDA model of an episode, the number of viewers from Wikipedia and the rating from IMDb. The LDA model is a generative probabilistic model of a collection of documents made up of words. Method In this prescriptive research, the case study method was used, and its results were analyzed using a quantitative approach. Summary of Results With the features of each episode, the model that performed the best to predict the rating was Catboost due to a similar mean squared error of the KNN model but a better standard deviation during the test phase. It was possible to predict IMDb ratings with an acceptable root mean squared error of 0.55.

Nontrivial class of anisotropic compact stellar model in Rastall gravity ; We investigated Rastall gravity, for an anisotropic star with a static spherical symmetry, whereas the mattergeometry coupling as assumed in Rastall Theory RT is expected to play a crucial role in differentiating RT from General Relativity GR. Indeed, all the obtained results confirm that RT is not equivalent to GR, however, it produces the same amount of anisotropy as GR for static spherically symmetric stellar models. We used the observational constraints on the mass and the radius of the pulsar textitHer X1 to determine the model parameters confirming the physical viability of the model. We found that the mattergeometry coupling in RT allows slightly less size than GR for a given mass. We confirmed the model viability via other twenty pulsars' observations. Utilizing the strong energy condition we determined an upper bound on compactness Utextmaxsim 0.603, in agreement with the Buchdahl limit, whereas Rastall parameter epsilon0.1. For a surface density compatible with a neutron core at nuclear saturation density, the massradius curve allows masses up to 3.53 Modot. We note that there is no equation of state is assumed, however, the model fits well with a linear behavior. We split the twenty pulsars into four groups according to the boundary densities. Three groups are compatible with neutron cores while one group fits perfectly with higher boundary density 8times 1014 gcm3 which suggests that those pulsars may have quarkgluon cores.

Automatic Dynamic Relevance Determination for Gaussian process regression with highdimensional functional inputs ; In the context of Gaussian process regression with functional inputs, it is common to treat the input as a vector. The parameter space becomes prohibitively complex as the number of functional points increases, effectively becoming a hindrance for automatic relevance determination in highdimensional problems. Generalizing a framework for timevarying inputs, we introduce the asymmetric Laplace functional weight ALF a flexible, parametric function that drives predictive relevance over the index space. Automatic dynamic relevance determination ADRD is achieved with three unknowns per input variable and enforces smoothness over the index space. Additionally, we discuss a screening technique to assess under complete absence of prior and model information whether ADRD is reasonably consistent with the data. Such tool may serve for exploratory analyses and model diagnostics. ADRD is applied to remote sensing data and predictions are generated in response to atmospheric functional inputs. Fully Bayesian estimation is carried out to identify relevant regions of the functional input space. Validation is performed to benchmark against traditional vectorinput model specifications. We find that ADRD outperforms models with input dimension reduction via functional principal component analysis. Furthermore, the predictive power is comparable to highdimensional models, in terms of both mean prediction and uncertainty, with 10 times fewer tuning parameters. Enforcing smoothness on the predictive relevance profile rules out erratic patterns associated with vectorinput models.

MultiModal Experience Inspired AI Creation ; AI creation, such as poem or lyrics generation, has attracted increasing attention from both industry and academic communities, with many promising models proposed in the past few years. Existing methods usually estimate the outputs based on single and independent visual or textual information. However, in reality, humans usually make creations according to their experiences, which may involve different modalities and be sequentially correlated. To model such human capabilities, in this paper, we define and solve a novel AI creation problem based on human experiences. More specifically, we study how to generate texts based on sequential multimodal information. Compared with the previous works, this task is much more difficult because the designed model has to well understand and adapt the semantics among different modalities and effectively convert them into the output in a sequential manner. To alleviate these difficulties, we firstly design a multichannel sequencetosequence architecture equipped with a multimodal attention network. For more effective optimization, we then propose a curriculum negative sampling strategy tailored for the sequential inputs. To benchmark this problem and demonstrate the effectiveness of our model, we manually labeled a new multimodal experience dataset. With this dataset, we conduct extensive experiments by comparing our model with a series of representative baselines, where we can demonstrate significant improvements in our model based on both automatic and humancentered metrics. The code and data are available at urlhttpsgithub.comAman4RealMMTG.

CGANECT Tomography Image Reconstruction from Electrical Capacitance Measurements Using CGANs ; Due to the rapid growth of Electrical Capacitance Tomography ECT applications in several industrial fields, there is a crucial need for developing high quality, yet fast, methodologies of image reconstruction from raw capacitance measurements. Deep learning, as an effective nonlinear mapping tool for complicated functions, has been going viral in many fields including electrical tomography. In this paper, we propose a Conditional Generative Adversarial Network CGAN model for reconstructing ECT images from capacitance measurements. The initial image of the CGAN model is constructed from the capacitance measurement. To our knowledge, this is the first time to represent the capacitance measurements in an image form. We have created a new massive ECT dataset of 320K synthetic image measurements pairs for training, and testing the proposed model. The feasibility and generalization ability of the proposed CGANECT model are evaluated using testing dataset, contaminated data and flow patterns that are not exposed to the model during the training phase. The evaluation results prove that the proposed CGANECT model can efficiently create more accurate ECT images than traditional and other deep learningbased image reconstruction algorithms. CGANECT achieved an average image correlation coefficient of more than 99.3 and an average relative image error about 0.07.

Sparsityguided Network Design for Frame Interpolation ; DNNbased frame interpolation, which generates intermediate frames from two consecutive frames, is often dependent on model architectures with a large number of features, preventing their deployment on systems with limited resources, such as mobile devices. We present a compressiondriven network design for frame interpolation that leverages model pruning through sparsityinducing optimization to greatly reduce the model size while attaining higher performance. Concretely, we begin by compressing the recently proposed AdaCoF model and demonstrating that a 10 times compressed AdaCoF performs similarly to its original counterpart, where different strategies for using layerwise sparsity information as a guide are comprehensively investigated under a variety of hyperparameter settings. We then enhance this compressed model by introducing a multiresolution warping module, which improves visual consistency with multilevel details. As a result, we achieve a considerable performance gain with a quarter of the size of the original AdaCoF. In addition, our model performs favorably against other stateoftheart approaches on a wide variety of datasets. We note that the suggested compressiondriven framework is generic and can be easily transferred to other DNNbased frame interpolation algorithms. The source code is available at httpsgithub.comtding1CDFI.

OPAL OntologyAware Pretrained Language Model for EndtoEnd TaskOriented Dialogue ; This paper presents an ontologyaware pretrained language model OPAL for endtoend taskoriented dialogue TOD. Unlike chitchat dialogue models, taskoriented dialogue models fulfill at least two taskspecific modules dialogue state tracker DST and response generator RG. The dialogue state consists of the domainslotvalue triples, which are regarded as the user's constraints to search the domainrelated databases. The largescale taskoriented dialogue data with the annotated structured dialogue state usually are inaccessible. It prevents the development of the pretrained language model for the taskoriented dialogue. We propose a simple yet effective pretraining method to alleviate this problem, which consists of two pretraining phases. The first phase is to pretrain on largescale contextual text data, where the structured information of the text is extracted by the information extracting tool. To bridge the gap between the pretraining method and downstream tasks, we design two pretraining tasks ontologylike triple recovery and nexttext generation, which simulates the DST and RG, respectively. The second phase is to finetune the pretrained model on the TOD data. The experimental results show that our proposed method achieves an exciting boost and get competitive performance even without any TOD data on CamRest676 and MultiWOZ benchmarks.

Correlation Information Bottleneck Towards Adapting Pretrained Multimodal Models for Robust Visual Question Answering ; Benefiting from largescale pretrained vision language models VLMs, the performance of visual question answering VQA has approached human oracles. However, finetuning such models on limited data often suffers from overfitting and poor generalization issues, leading to a lack of model robustness. In this paper, we aim to improve input robustness from an information bottleneck perspective when adapting pretrained VLMs to the downstream VQA task. Input robustness refers to the ability of models to defend against visual and linguistic input variations, as well as shortcut learning involved in inputs. Generally, the representations obtained by pretrained VLMs inevitably contain irrelevant and redundant information for a specific downstream task, resulting in statistically spurious correlations and insensitivity to input variations. To encourage representations to converge to a minimal sufficient statistic in multimodal learning, we propose Correlation Information Bottleneck CIB, which seeks a tradeoff between compression and redundancy in representations by minimizing the mutual information MI between inputs and representations while maximizing the MI between outputs and representations. Moreover, we derive a tight theoretical upper bound for the mutual information between multimodal inputs and representations, incorporating different internal correlations that guide models to learn more robust representations and facilitate modality alignment. Extensive experiments consistently demonstrate the effectiveness and superiority of the proposed CIB in terms of input robustness and accuracy.

Answering Numerical Reasoning Questions in TableText Hybrid Contents with Graphbased Encoder and Treebased Decoder ; In the realworld question answering scenarios, hybrid form combining both tabular and textual contents has attracted more and more attention, among which numerical reasoning problem is one of the most typical and challenging problems. Existing methods usually adopt encoderdecoder framework to represent hybrid contents and generate answers. However, it can not capture the rich relationship among numerical value, table schema, and text information on the encoder side. The decoder uses a simple predefined operator classifier which is not flexible enough to handle numerical reasoning processes with diverse expressions. To address these problems, this paper proposes a textbfRelational textbfGraph enhanced textbfHybrid tabletext textbfNumerical reasoning model with textbfTree decoder textbfRegHNT. It models the numerical question answering over tabletext hybrid contents as an expression tree generation task. Moreover, we propose a novel relational graph modeling method, which models alignment between questions, tables, and paragraphs. We validated our model on the publicly available tabletext hybrid QA benchmark TATQA. The proposed RegHNT significantly outperform the baseline model and achieve stateoftheart results. We openly released the source code and data at httpsgithub.comlfy79001RegHNT 20220505.

PointCAT Contrastive Adversarial Training for Robust Point Cloud Recognition ; Notwithstanding the prominent performance achieved in various applications, point cloud recognition models have often suffered from natural corruptions and adversarial perturbations. In this paper, we delve into boosting the general robustness of point cloud recognition models and propose PointCloud Contrastive Adversarial Training PointCAT. The main intuition of PointCAT is encouraging the target recognition model to narrow the decision gap between clean point clouds and corrupted point clouds. Specifically, we leverage a supervised contrastive loss to facilitate the alignment and uniformity of the hypersphere features extracted by the recognition model, and design a pair of centralizing losses with the dynamic prototype guidance to avoid these features deviating from their belonging category clusters. To provide the more challenging corrupted point clouds, we adversarially train a noise generator along with the recognition model from the scratch, instead of using gradientbased attack as the inner loop like previous adversarial training methods. Comprehensive experiments show that the proposed PointCAT outperforms the baseline methods and dramatically boosts the robustness of different point cloud recognition models, under a variety of corruptions including isotropic point noises, the LiDAR simulated noises, random point dropping and adversarial perturbations.

Autoregressive Transformers for DataDriven SpatioTemporal Learning of Turbulent Flows ; A convolutional encoderdecoderbased transformer model is proposed for autoregressively training on spatiotemporal data of turbulent flows. The prediction of future fluid flow fields is based on the previously predicted fluid flow field to ensure longterm predictions without diverging. A combination of convolutional neural networks and transformer architecture is utilized to handle both the spatial and temporal dimensions of the data. To assess the performance of the model, a priori assessments are conducted, and significant agreements are found with the ground truth data. The a posteriori predictions, which are generated after a considerable number of simulation steps, exhibit predicted variances. The autoregressive training and prediction of a posteriori states are deemed crucial steps towards the development of more complex datadriven turbulence models and simulations. The highly nonlinear and chaotic dynamics of turbulent flows can be handled by the proposed model, and accurate predictions over long time horizons can be generated. Overall, the potential of using deep learning techniques to improve the accuracy and efficiency of turbulence modeling and simulation is demonstrated by this approach. The proposed model can be further optimized and extended to incorporate additional physics and boundary conditions, paving the way for more realistic simulations of complex fluid dynamics.

Predicting DrugDrug Interactions using Deep Generative Models on Graphs ; Latent representations of drugs and their targets produced by contemporary graph autoencoderbased models have proved useful in predicting many types of nodepair interactions on large networks, including drugdrug, drugtarget, and targettarget interactions. However, most existing approaches model the node's latent spaces in which node distributions are rigid and disjoint; these limitations hinder the methods from generating new links among pairs of nodes. In this paper, we present the effectiveness of variational graph autoencoders VGAE in modeling latent node representations on multimodal networks. Our approach can produce flexible latent spaces for each node type of the multimodal graph; the embeddings are used later for predicting links among node pairs under different edge types. To further enhance the models' performance, we suggest a new method that concatenates Morgan fingerprints, which capture the molecular structures of each drug, with their latent embeddings before preceding them to the decoding stage for link prediction. Our proposed model shows competitive results on two multimodal networks 1 a multigraph consisting of drug and protein nodes, and 2 a multigraph consisting of drug and cell line nodes. Our source code is publicly available at httpsgithub.comHySonLabdruginteractions.

Black hole mass and spin measurements through the Relativistic Precession Model XTE J1859226 ; The Xray light curves of accreting black holes and neutron stars in binary systems show various types of quasiperiodic oscillations QPOs, the origin of which is still debated. The Relativistic Precession Model identifies the QPO frequencies with fundamental time scales from General Relativity, and has been proposed as a possible explanation of certain types of such oscillations. Under specific conditions i.e., the detection of a particular QPOs triplet such a model can be used to obtain selfconsistent measurements of the mass and spin of the compact object. So far this has been possible only in the black hole binary GRO J165540. In the RXTEPCA data from the 19992000 outburst of the black hole transient XTE J1859226 we found a QPO triplet, and used the the Relativistic Precession Model to obtain highprecision measurements of the black hole mass and spin M 7.850.46 Msun, a 0.1490.005 the former being consistent with the most recent dynamical mass determination from optical measurements. Similarly to what has been already observed in other black hole systems, the frequencies of the QPOs and broadband noise components match the general relativistic frequencies of particle motion close to the compact object predicted by the model. Our findings confirm previous results and further support the validity of the Relativistic Precession Model, which is the only electromagneticmeasurementbased method that so far has consistently yielded spins close to those from the gravitational waves produced by merging binary black holes.

PACT PerceptionAction Causal Transformer for Autoregressive Robotics PreTraining ; Robotics has long been a field riddled with complex systems architectures whose modules and connections, whether traditional or learningbased, require significant human expertise and prior knowledge. Inspired by large pretrained language models, this work introduces a paradigm for pretraining a general purpose representation that can serve as a starting point for multiple tasks on a given robot. We present the PerceptionAction Causal Transformer PACT, a generative transformerbased architecture that aims to build representations directly from robot data in a selfsupervised fashion. Through autoregressive prediction of states and actions over time, our model implicitly encodes dynamics and behaviors for a particular robot. Our experimental evaluation focuses on the domain of mobile agents, where we show that this robotspecific representation can function as a single starting point to achieve distinct tasks such as safe navigation, localization and mapping. We evaluate two form factors a wheeled robot that uses a LiDAR sensor as perception input MuSHR, and a simulated agent that uses firstperson RGB images Habitat. We show that finetuning small taskspecific networks on top of the larger pretrained model results in significantly better performance compared to training a single model from scratch for all tasks simultaneously, and comparable performance to training a separate large model for each task independently. By sharing a common goodquality representation across tasks we can lower overall model capacity and speed up the realtime deployment of such systems.

MeWEHV Mel and Wave Embeddings for Human Voice Tasks ; A recent trend in speech processing is the use of embeddings created through machine learning models trained on a specific task with large datasets. By leveraging the knowledge already acquired, these models can be reused in new tasks where the amount of available data is small. This paper proposes a pipeline to create a new model, called Mel and Wave Embeddings for Human Voice Tasks MeWEHV, capable of generating robust embeddings for speech processing. MeWEHV combines the embeddings generated by a pretrained raw audio waveform encoder model, and deep features extracted from Mel Frequency Cepstral Coefficients MFCCs using Convolutional Neural Networks CNNs. We evaluate the performance of MeWEHV on three tasks speaker, language, and accent identification. For the first one, we use the VoxCeleb1 dataset and present YouSpeakers204, a new and publicly available dataset for English speaker identification that contains 19607 audio clips from 204 persons speaking in six different accents, allowing other researchers to work with a very balanced dataset, and to create new models that are robust to multiple accents. For evaluating the language identification task, we use the VoxForge and Common Language datasets. Finally, for accent identification, we use the Latin American Spanish Corpora LASC and Common Voice datasets. Our approach allows a significant increase in the performance of stateoftheart models on all the tested datasets, with a low additional computational cost.

Masked MultiStep Multivariate Time Series Forecasting with Future Information ; In this paper, we introduce Masked MultiStep Multivariate Forecasting MMMF, a novel and general selfsupervised learning framework for time series forecasting with known future information. In many realworld forecasting scenarios, some future information is known, e.g., the weather information when making a shorttomidterm electricity demand forecast, or the oil price forecasts when making an airplane departure forecast. Existing machine learning forecasting frameworks can be categorized into 1 samplebased approaches where each forecast is made independently, and 2 time series regression approaches where the future information is not fully incorporated. To overcome the limitations of existing approaches, we propose MMMF, a framework to train any neural network model capable of generating a sequence of outputs, that combines both the temporal information from the past and the known information about the future to make better predictions. Experiments are performed on two realworld datasets for 1 midterm electricity demand forecasting, and 2 twomonth ahead flight departures forecasting. They show that the proposed MMMF framework outperforms not only samplebased methods but also existing time series forecasting models with the exact same base models. Furthermore, once a neural network model is trained with MMMF, its inference speed is similar to that of the same model trained with traditional regression formulations, thus making MMMF a better alternative to existing regressiontrained time series forecasting models if there is some available future information.

Learning by Distilling Context ; Language models significantly benefit from context tokens, such as prompts or scratchpads. They perform better when prompted with informative instructions, and they acquire new reasoning capabilities by generating a scratchpad before predicting the final answers. However, they do not textitinternalize these performance gains, which disappear when the context tokens are gone. Our work proposes to apply context distillation so that a language model can improve itself by internalizing these gains. Concretely, given a synthetic unlabeled input for the target task, we condition the model on instructions taskinput'' to predict scratchpad final answer''; then we finetune the same model to predict its own final answer'' conditioned on the taskinput'', without seeing the instructions'' or using the scratchpad''. We show that context distillation is a general method to train language models, and it can effectively internalize 3 types of training signals. First, it can internalize abstract task instructions and explanations, so we can iteratively update the model parameters with new instructions and overwrite old ones. Second, it can internalize stepbystep reasoning for complex tasks e.g., 8digit addition, and such a newly acquired capability proves to be useful for other downstream tasks. Finally, it can internalize concrete training examples, and it outperforms directly learning with gradient descent by 9 on the SPIDER TexttoSQL dataset; furthermore, combining context distillation operations can internalize more training examples than the context window size allows.

Domain Discrepancy Aware Distillation for Model Aggregation in Federated Learning ; Knowledge distillation has recently become popular as a method of model aggregation on the server for federated learning. It is generally assumed that there are abundant public unlabeled data on the server. However, in reality, there exists a domain discrepancy between the datasets of the server domain and a client domain, which limits the performance of knowledge distillation. How to improve the aggregation under such a domain discrepancy setting is still an open problem. In this paper, we first analyze the generalization bound of the aggregation model produced from knowledge distillation for the client domains, and then describe two challenges, servertoclient discrepancy and clienttoclient discrepancy, brought to the aggregation model by the domain discrepancies. Following our analysis, we propose an adaptive knowledge aggregation algorithm FedD3A based on domain discrepancy aware distillation to lower the bound. FedD3A performs adaptive weighting at the sample level in each round of FL. For each sample in the server domain, only the client models of its similar domains will be selected for playing the teacher role. To achieve this, we show that the discrepancy between the serverside sample and the client domain can be approximately measured using a subspace projection matrix calculated on each client without accessing its raw data. The server can thus leverage the projection matrices from multiple clients to assign weights to the corresponding teacher models for each serverside sample. We validate FedD3A on two popular crossdomain datasets and show that it outperforms the compared competitors in both crosssilo and crossdevice FL settings.

Supervised and Unsupervised Learning of Audio Representations for Music Understanding ; In this work, we provide a broad comparative analysis of strategies for pretraining audio understanding models for several tasks in the music domain, including labelling of genre, era, origin, mood, instrumentation, key, pitch, vocal characteristics, tempo and sonority. Specifically, we explore how the domain of pretraining datasets music or generic audio and the pretraining methodology supervised or unsupervised affects the adequacy of the resulting audio embeddings for downstream tasks. We show that models trained via supervised learning on largescale expertannotated music datasets achieve stateoftheart performance in a wide range of music labelling tasks, each with novel content and vocabularies. This can be done in an efficient manner with models containing less than 100 million parameters that require no finetuning or reparameterization for downstream tasks, making this approach practical for industryscale audio catalogs. Within the class of unsupervised learning strategies, we show that the domain of the training dataset can significantly impact the performance of representations learned by the model. We find that restricting the domain of the pretraining dataset to music allows for training with smaller batch sizes while achieving stateoftheart in unsupervised learning and in some cases, supervised learning for music understanding. We also corroborate that, while achieving stateoftheart performance on many tasks, supervised learning can cause models to specialize to the supervised information provided, somewhat compromising a model's generality.

On Designing Day Ahead and Same Day Ridership Level Prediction Models for CityScale Transit Networks Using Noisy APC Data ; The ability to accurately predict public transit ridership demand benefits passengers and transit agencies. Agencies will be able to reallocate buses to handle under or overutilized bus routes, improving resource utilization, and passengers will be able to adjust and plan their schedules to avoid overcrowded buses and maintain a certain level of comfort. However, accurately predicting occupancy is a nontrivial task. Various reasons such as heterogeneity, evolving ridership patterns, exogenous events like weather, and other stochastic variables, make the task much more challenging. With the progress of big data, transit authorities now have access to realtime passenger occupancy information for their vehicles. The amount of data generated is staggering. While there is no shortage in data, it must still be cleaned, processed, augmented, and merged before any useful information can be generated. In this paper, we propose the use and fusion of data from multiple sources, cleaned, processed, and merged together, for use in training machine learning models to predict transit ridership. We use data that spans a 2year period 20202022 incorporating transit, weather, traffic, and calendar data. The resulting data, which equates to 17 million observations, is used to train separate models for the trip and stop level prediction. We evaluate our approach on realworld transit data provided by the public transit agency of Nashville, TN. We demonstrate that the trip level model based on Xgboost and the stop level model based on LSTM outperform the baseline statistical model across the entire transit service day.

On Text Style Transfer via Style Masked Language Models ; Text Style Transfer TST is performable through approaches such as latent space disentanglement, cycleconsistency losses, prototype editing etc. The prototype editing approach, which is known to be quite successful in TST, involves two key phases a Masking of source styleassociated tokens and b Reconstruction of this sourcestyle masked sentence conditioned with the target style. We follow a similar transduction method, in which we transpose the more difficult direct source to target TST task to a simpler StyleMasked Language Model SMLM Task, wherein, similar to BERT citebert, the goal of our model is now to reconstruct the source sentence from its stylemasked version. We arrive at the SMLM mechanism naturally by formulating prototype editing transduction methods in a probabilistic framework, where TST resolves into estimating a hypothetical parallel dataset from a partially observed parallel dataset, wherein each domain is assumed to have a common latent stylemasked prior. To generate this stylemasked prior, we use Explainable Attention as our choice of attribution for a more precise stylemasking step and also introduce a costeffective and accurate AttributionSurplus method of determining the position of masks from any arbitrary attribution model in O1 time. We empirically show that this nongenerational approach well suites the content preserving criteria for a task like TST, even for a complex style like Discourse Manipulation. Our model, the Style MLM, outperforms strong TST baselines and is on par with stateoftheart TST models, which use complex architectures and orders of more parameters.

MAPL ParameterEfficient Adaptation of Unimodal PreTrained Models for VisionLanguage FewShot Prompting ; Large pretrained models have proved to be remarkable zero and promptbased fewshot learners in unimodal vision and language tasks. We propose MAPL, a simple and parameterefficient method that reuses frozen pretrained unimodal models and leverages their strong generalization capabilities in multimodal visionlanguage VL settings. MAPL learns a lightweight mapping between the representation spaces of unimodal models using aligned imagetext data, and can generalize to unseen VL tasks from just a few incontext examples. The small number of trainable parameters makes MAPL effective at lowdata and indomain learning. Moreover, MAPL's modularity enables easy extension to other pretrained models. Extensive experiments on several visual question answering and image captioning benchmarks show that MAPL achieves superior or competitive performance compared to similar methods while training orders of magnitude fewer parameters. MAPL can be trained in just a few hours using modest computational resources and public datasets. We release our code and pretrained model weights at httpsgithub.commairlabmapl.

LeMoN Lens Modelling with Neural networks I. Automated modelling of strong gravitational lenses with Bayesian Neural Networks ; The unprecedented number of gravitational lenses expected from newgeneration facilities such as the ESA Euclid telescope and the Vera Rubin Observatory makes it crucial to rethink our classical approach to lensmodelling. In this paper, we present LeMoN Lens Modelling with Neural networks a new machinelearning algorithm able to analyse hundreds of thousands of gravitational lenses in a reasonable amount of time. The algorithm is based on a Bayesian Neural Network a new generation of neural networks able to associate a reliable confidence interval to each predicted parameter. We train the algorithm to predict the three main parameters of the Singular Isothermal Ellipsoid model the Einstein radius and the two components of the ellipticity by employing two simulated datasets built to resemble the imaging capabilities of the Hubble Space Telescope and the forthcoming Euclid satellite. In this work, we assess the accuracy of the algorithm and the reliability of the estimated uncertainties by applying the network to several simulated datasets of 10.000 images each. We obtain accuracies comparable to previous studies present in the current literature and an average modelling time of just 0.5s per lens. Finally, we apply the LeMoN algorithm to a pilot dataset of real lenses observed with HST during the SLACS program, obtaining unbiased estimates of their SIE parameters. The code is publicly available on GitHub httpsgithub.comfabgentileLeMoN.

Multilevel Data Representation For Training Deep Helmholtz Machines ; A vast majority of the current research in the field of Machine Learning is done using algorithms with strong arguments pointing to their biological implausibility such as Backpropagation, deviating the field's focus from understanding its original organic inspiration to a compulsive search for optimal performance. Yet, there have been a few proposed models that respect most of the biological constraints present in the human brain and are valid candidates for mimicking some of its properties and mechanisms. In this paper, we will focus on guiding the learning of a biologically plausible generative model called the Helmholtz Machine in complex search spaces using a heuristic based on the Human Image Perception mechanism. We hypothesize that this model's learning algorithm is not fit for Deep Networks due to its Hebbianlike local update rule, rendering it incapable of taking full advantage of the compositional properties that multilayer networks provide. We propose to overcome this problem, by providing the network's hidden layers with visual queues at different resolutions using a Multilevel Data representation. The results on several image datasets showed the model was able to not only obtain better overall quality but also a wider diversity in the generated images, corroborating our intuition that using our proposed heuristic allows the model to take more advantage of the network's depth growth. More importantly, they show the unexplored possibilities underlying braininspired models and techniques.

Exploiting Features and Logits in Heterogeneous Federated Learning ; Due to the rapid growth of IoT and artificial intelligence, deploying neural networks on IoT devices is becoming increasingly crucial for edge intelligence. Federated learning FL facilitates the management of edge devices to collaboratively train a shared model while maintaining training data local and private. However, a general assumption in FL is that all edge devices are trained on the same machine learning model, which may be impractical considering diverse device capabilities. For instance, less capable devices may slow down the updating process because they struggle to handle large models appropriate for ordinary devices. In this paper, we propose a novel datafree FL method that supports heterogeneous client models by managing features and logits, called Felo; and its extension with a conditional VAE deployed in the server, called Velo. Felo averages the midlevel features and logits from the clients at the server based on their class labels to provide the average features and logits, which are utilized for further training the client models. Unlike Felo, the server has a conditional VAE in Velo, which is used for training midlevel features and generating synthetic features according to the labels. The clients optimize their models based on the synthetic features and the average logits. We conduct experiments on two datasets and show satisfactory performances of our methods compared with the stateoftheart methods.

Delivery by Drones with Arbitrary Energy Consumption Models A New Formulation Approach ; This paper presents a new approach for formulating the delivery problem by drones with general energy consumption models where the drones visit a set of places to deliver parcels to customers. Drones can perform multiple trips that start and end at a central depot while visiting several customers along their paths. The problem determines the routing and scheduling decisions of the drones in order to minimize the total transportation cost of serving customers. For the first time, the new formulation approach enables us to use the best available energy consumption model without the need of any extra approximations. Though the approach works in a very general setting including nonconvex energy consumption models, it is also computationally efficient as the resulting optimization model has a linear relaxation. A numerical study on 255 benchmark instances with up to 50 customers and a specific energy function indicate that all the instances can be solved 20 times faster on average using the new formulation when compared to the best existing branchandcut algorithm. All the 15 benchmark instances with 50 customers are solved exactly, whereas none of them has been solved optimally before. Moreover, new instances with up to 150 customers are solved with small error bounds within a few hours. The new approach can be simply applied to consider the extra energy required when a drone needs to continue hovering until opening the delivery time window. It can also be applied to the case where the flight time is dependent on the drone's payload weight. Owing to the flexibility of the new approach, these challenging extensions are formulated as linear optimization models for the first time.

Characterizing Intrinsic Compositionality in Transformers with Tree Projections ; When trained on language data, do transformers learn some arbitrary computation that utilizes the full capacity of the architecture or do they learn a simpler, treelike computation, hypothesized to underlie compositional meaning systems like human languages There is an apparent tension between compositional accounts of human language understanding, which are based on a restricted bottomup computational process, and the enormous success of neural models like transformers, which can route information arbitrarily between different parts of their input. One possibility is that these models, while extremely flexible in principle, in practice learn to interpret language hierarchically, ultimately building sentence representations close to those predictable by a bottomup, treestructured model. To evaluate this possibility, we describe an unsupervised and parameterfree method to emphfunctionally project the behavior of any transformer into the space of treestructured networks. Given an input sentence, we produce a binary tree that approximates the transformer's representationbuilding process and a score that captures how treelike the transformer's behavior is on the input. While calculation of this score does not require training any additional models, it provably upperbounds the fit between a transformer and any treestructured approximation. Using this method, we show that transformers for three different tasks become more treelike over the course of training, in some cases unsupervisedly recovering the same trees as supervised parsers. These trees, in turn, are predictive of model behavior, with more treelike models generalizing better on tests of compositional generalization.

Harnessing the Power of Explanations for Incremental Training A LIMEBased Approach ; Explainability of neural network prediction is essential to understand feature importance and gain interpretable insight into neural network performance. However, explanations of neural network outcomes are mostly limited to visualization, and there is scarce work that looks to use these explanations as feedback to improve model performance. In this work, model explanations are fed back to the feedforward training to help the model generalize better. To this extent, a custom weighted loss where the weights are generated by considering the Euclidean distances between true LIME Local Interpretable ModelAgnostic Explanations explanations and modelpredicted LIME explanations is proposed. Also, in practical training scenarios, developing a solution that can help the model learn sequentially without losing information on previous data distribution is imperative due to the unavailability of all the training data at once. Thus, the framework incorporates the custom weighted loss with Elastic Weight Consolidation EWC to maintain performance in sequential testing sets. The proposed custom training procedure results in a consistent enhancement of accuracy ranging from 0.5 to 1.5 throughout all phases of the incremental learning setup compared to traditional lossbased training methods for the keyword spotting task using the Google Speech Commands dataset.

Fast Noise Removal in Hyperspectral Images via Representative Coefficient Total Variation ; Mining structural priors in data is a widely recognized technique for hyperspectral image HSI denoising tasks, whose typical ways include modelbased methods and databased methods. The modelbased methods have good generalization ability, while the runtime cannot meet the fast processing requirements of the practical situations due to the large size of an HSI data mathbfX in mathbbRMNtimes B. For the databased methods, they perform very fast on new test data once they have been trained. However, their generalization ability is always insufficient. In this paper, we propose a fast modelbased HSI denoising approach. Specifically, we propose a novel regularizer named Representative Coefficient Total Variation RCTV to simultaneously characterize the low rank and local smooth properties. The RCTV regularizer is proposed based on the observation that the representative coefficient matrix mathbfUinmathbbRMNtimes R Rll B obtained by orthogonally transforming the original HSI mathbfX can inherit the strong localsmooth prior of mathbfX. Since RB is very small, the HSI denoising model based on the RCTV regularizer has lower time complexity. Additionally, we find that the representative coefficient matrix mathbfU is robust to noise, and thus the RCTV regularizer can somewhat promote the robustness of the HSI denoising model. Extensive experiments on mixed noise removal demonstrate the superiority of the proposed method both in denoising performance and denoising speed compared with other stateoftheart methods. Remarkably, the denoising speed of our proposed method outperforms all the modelbased techniques and is comparable with the deep learningbased approaches.

Integrated ParameterEfficient Tuning for GeneralPurpose Audio Models ; The advent of hyperscale and generalpurpose pretrained models is shifting the paradigm of building taskspecific models for target tasks. In the field of audio research, taskagnostic pretrained models with high transferability and adaptability have achieved stateoftheart performances through finetuning for downstream tasks. Nevertheless, retraining all the parameters of these massive models entails an enormous amount of time and cost, along with a huge carbon footprint. To overcome these limitations, the present study explores and applies efficient transfer learning methods in the audio domain. We also propose an integrated parameterefficient tuning IPET framework by aggregating the embedding prompt a promptbased learning approach, and the adapter an effective transfer learning method. We demonstrate the efficacy of the proposed framework using two backbone pretrained audio models with different characteristics the audio spectrogram transformer and wav2vec 2.0. The proposed IPET framework exhibits remarkable performance compared to finetuning method with fewer trainable parameters in four downstream tasks sound event classification, music genre classification, keyword spotting, and speaker verification. Furthermore, the authors identify and analyze the shortcomings of the IPET framework, providing lessons and research directions for parameter efficient tuning in the audio domain.

Domainincremental Cardiac Image Segmentation with Styleoriented Replay and Domainsensitive Feature Whitening ; Contemporary methods have shown promising results on cardiac image segmentation, but merely in static learning, i.e., optimizing the network once for all, ignoring potential needs for model updating. In realworld scenarios, new data continues to be gathered from multiple institutions over time and new demands keep growing to pursue more satisfying performance. The desired model should incrementally learn from each incoming dataset and progressively update with improved functionality as time goes by. As the datasets sequentially delivered from multiple sites are normally heterogenous with domain discrepancy, each updated model should not catastrophically forget previously learned domains while well generalizing to currently arrived domains or even unseen domains. In medical scenarios, this is particularly challenging as accessing or storing past data is commonly not allowed due to data privacy. To this end, we propose a novel domainincremental learning framework to recover past domain inputs first and then regularly replay them during model optimization. Particularly, we first present a styleoriented replay module to enable structurerealistic and memoryefficient reproduction of past data, and then incorporate the replayed past data to jointly optimize the model with current data to alleviate catastrophic forgetting. During optimization, we additionally perform domainsensitive feature whitening to suppress model's dependency on features that are sensitive to domain changes e.g., domaindistinctive style features to assist domaininvariant feature exploration and gradually improve the generalization performance of the network. We have extensively evaluated our approach with the MMs Dataset in singledomain and compounddomain incremental learning settings with improved performance over other comparison approaches.

Neural Network Approaches for Data Estimation in Unique Word OFDM Systems ; Data estimation is conducted with modelbased estimation methods since the beginning of digital communications. However, motivated by the growing success of machine learning, current research focuses on replacing modelbased data estimation methods by datadriven approaches, mainly neural networks NNs. In this work, we particularly investigate the incorporation of existing model knowledge into datadriven approaches, which is expected to lead to complexity reduction and or performance enhancement. We describe three different options, namely modelinspired'' preprocessing, choosing an NN architecture motivated by the properties of the underlying communication system, and inferring the layer structure of an NN with the help of model knowledge. Most of the current publications on NNbased data estimation deal with general multipleinput multipleoutput communication MIMO systems. In this work, we investigate NNbased data estimation for socalled unique word orthogonal frequency division multiplexing UWOFDM systems. We highlight differences between UWOFDM systems and general MIMO systems one has to be aware of when using NNs for data estimation, and we introduce measures for successful utilization of NNbased data estimators in UWOFDM systems. Further, we investigate the use of NNs for data estimation when channel coded data transmission is conducted, and we present adaptions to be made, such that NNbased data estimators provide satisfying performance for this case. We compare the presented NNs concerning achieved bit error ratio performance and computational complexity, we show the peculiar distributions of their data estimates, and we also point out their downsides compared to modelbased equalizers.

Towards an objective characterization of an individual's facial movements using SelfSupervised PersonSpecificModels ; Disentangling facial movements from other facial characteristics, particularly from facial identity, remains a challenging task, as facial movements display great variation between individuals. In this paper, we aim to characterize individualspecific facial movements. We present a novel training approach to learn facial movements independently of other facial characteristics, focusing on each individual separately. We propose selfsupervised PersonSpecific Models PSMs, in which one model per individual can learn to extract an embedding of the facial movements independently of the person's identity and other structural facial characteristics from unlabeled facial video. These models are trained using encoderdecoderlike architectures. We provide quantitative and qualitative evidence that a PSM learns a meaningful facial embedding that discovers finegrained movements otherwise not characterized by a General Model GM, which is trained across individuals and characterizes general patterns of facial movements. We present quantitative and qualitative evidence that this approach is easily scalable and generalizable for new individuals facial movements knowledge learned on a person can quickly and effectively be transferred to a new person. Lastly, we propose a novel PSM using curriculum temporal learning to leverage the temporal contiguity between video frames. Our code, analysis details, and all pretrained models are available in Github and Supplementary Materials.

Progressive TreeStructured Prototype Network for EndtoEnd Image Captioning ; Studies of image captioning are shifting towards a trend of a fully endtoend paradigm by leveraging powerful visual pretrained models and transformerbased generation architecture for more flexible model training and faster inference speed. Stateoftheart approaches simply extract isolated concepts or attributes to assist description generation. However, such approaches do not consider the hierarchical semantic structure in the textual domain, which leads to an unpredictable mapping between visual representations and concept words. To this end, we propose a novel Progressive TreeStructured prototype Network dubbed PTSN, which is the first attempt to narrow down the scope of prediction words with appropriate semantics by modeling the hierarchical textual semantics. Specifically, we design a novel embedding method called treestructured prototype, producing a set of hierarchical representative embeddings which capture the hierarchical semantic structure in textual space. To utilize such treestructured prototypes into visual cognition, we also propose a progressive aggregation module to exploit semantic relationships within the image and prototypes. By applying our PTSN to the endtoend captioning framework, extensive experiments conducted on MSCOCO dataset show that our method achieves a new stateoftheart performance with 144.2 single model and 146.5 ensemble of 4 models CIDEr scores on Karpathy' split and 141.4 c5 and 143.9 c40 CIDEr scores on the official online test server. Trained models and source code have been released at httpsgithub.comNovaMindZPTSN.

Turning Silver into Gold Domain Adaptation with Noisy Labels for Wearable CardioRespiratory Fitness Prediction ; Deep learning models have shown great promise in various healthcare applications. However, most models are developed and validated on smallscale datasets, as collecting highquality goldstandard labels for health applications is often costly and timeconsuming. As a result, these models may suffer from overfitting and not generalize well to unseen data. At the same time, an extensive amount of data with imprecise labels silverstandard is starting to be generally available, as collected from inexpensive wearables like accelerometers and electrocardiography sensors. These currently underutilized datasets and labels can be leveraged to produce more accurate clinical models. In this work, we propose UDAMA, a novel model with two key components Unsupervised Domain Adaptation and Multidiscriminator Adversarial training, which leverage noisy data from source domain the silverstandard dataset to improve goldstandard modeling. We validate our framework on the challenging task of predicting labmeasured maximal oxygen consumption VO2max, the benchmark metric of cardiorespiratory fitness, using freeliving wearable sensor data from two cohort studies as inputs. Our experiments show that the proposed framework achieves the best performance of corr 0.665 pm 0.04, paving the way for accurate fitness estimation at scale.

LMAE Masked Autoencoders are Semantic Segmentation Datasets Augmenter ; Generating semantic segmentation datasets has consistently been laborious and timeconsuming, particularly in the context of large models or specialized domainsi.e. Medical Imaging or Remote Sensing. Specifically, large models necessitate a substantial volume of data, while datasets in professional domains frequently require the involvement of domain experts. Both scenarios are susceptible to inaccurate data labeling, which can significantly affect the ultimate performance of the trained model. This paper proposes a simple and effective label pixellevel completion method, textbfLabel Mask AutoEncoder LMAE, which fully uses the existing information in the label to generate the complete label. The proposed model are the first to apply the Mask AutoEncoder to downstream tasks. In detail, LMAE adopts the fusion strategy that stacks the label and the corresponding image, namely fuse map. Moreover, since some of the image information is lost when masking the fuse map, direct reconstruction may lead to poor performance. We proposed Image Patch Supplement algorithm to supplement the missing information during the maskreconstruct process, and empirically found that an average of 4.1 mIoU can be improved. We conducted a experiment to evaluate the efficacy of LMAE to complete the dataset. We employed a degraded Pascal VOC dataset and the degraded dataset enhanced by LMAE to train an identical conventional semantic segmentation model for the initial set of experiments. The results of these experiments demonstrate a performance enhancement of 13.5 in the model trained with the LMAEenhanced dataset compared to the unenhanced dataset.

Conditional diffusionbased microstructure reconstruction ; Microstructure reconstruction, a major component of inverse computational materials engineering, is currently advancing at an unprecedented rate. While various trainingbased and trainingfree approaches are developed, the majority of contributions are based on generative adversarial networks. In contrast, diffusion models constitute a more stable alternative, which have recently become the new state of the art and currently attract much attention. The present work investigates the applicability of diffusion models to the reconstruction of realworld microstructure data. For this purpose, a highly diverse and morphologically complex data set is created by combining and processing databases from the literature, where the reconstruction of realistic micrographs for a given material class demonstrates the ability of the model to capture these features. Furthermore, a fiber composite data set is used to validate the applicability of diffusion models to small data set sizes that can realistically be created by a single lab. The quality and diversity of the reconstructed microstructures is quantified by means of descriptorbased error metrics as well as the Fr'echet inception distance FID score. Although not present in the training data set, the generated samples are visually indistinguishable from real data to the untrained eye and various error metrics are computed. This demonstrates the utility of diffusion models in microstructure reconstruction and provides a basis for further extensions such as 2Dto3D reconstruction or application to multiscale modeling and structureproperty linkages.

The critical role of nuclear heating rates, thermalization efficiencies and opacities for kilonova modelling and parameter inference ; We present an improved version of the 3D Monte Carlo radiative transfer code POSSIS to model kilonovae from neutron star mergers, wherein nuclear heating rates, thermalization efficiencies and wavelengthdependent opacities depend on local properties of the ejecta and time. Using an axiallysymmetric twocomponent ejecta model, we explore how simplistic assumptions on heating rates, thermalization efficiencies and opacities often found in the literature affect kilonova spectra and light curves. Specifically, we compute five models one textttFIDUCIAL with an appropriate treatment of these three quantities, one textttSIMPLEHEAT with uniform heating rates throughout the ejecta, one textttSIMPLETHERM with a constant and uniform thermalization efficiency, one textttSIMPLEOPAC with grey opacities and one textttSIMPLEALL with all these three simplistic assumptions combined. We find that deviations from the textttFIDUCIAL model are of several sim110 magnitudes and are generally larger for the textttSIMPLEOPAC and textttSIMPLEALL compared to the textttSIMPLETHERM and textttSIMPLEHEAT models. The discrepancies generally increase from a faceon to an edgeon view of the system, from early to late epochs and from infrared to ultravioletoptical wavelengths. Our work indicates that kilonova studies using either of these simplistic assumptions ought to be treated with caution and that appropriate systematic uncertainties ought to be added to kilonova light curves when performing inference on ejecta parameters.

VRGNN Variational Relation Vector Graph Neural Network for Modeling both Homophily and Heterophily ; Graph Neural Networks GNNs have achieved remarkable success in diverse realworld applications. Traditional GNNs are designed based on homophily, which leads to poor performance under heterophily scenarios. Current solutions deal with heterophily mainly by mixing highorder neighbors or passing signed messages. However, mixing highorder neighbors destroys the original graph structure and passing signed messages utilizes an inflexible messagepassing mechanism, which is prone to producing unsatisfactory effects. To overcome the above problems, we propose a novel GNN model based on relation vector translation named Variational Relation Vector Graph Neural Network VRGNN. VRGNN models relation generation and graph aggregation into an endtoend model based on Variational AutoEncoder. The encoder utilizes the structure, feature and label to generate a proper relation vector. The decoder achieves superior node representation by incorporating the relation translation into the messagepassing framework. VRGNN can fully capture the homophily and heterophily between nodes due to the great flexibility of relation translation in modeling neighbor relationships. We conduct extensive experiments on eight realworld datasets with different homophilyheterophily properties to verify the effectiveness of our model. The experimental results show that VRGNN gains consistent and significant improvements against stateoftheart GNN methods under heterophily, and competitive performance under homophily.

Spontaneous nonHermiticity in the 21dimensional Thirring model ; Using the CornwallJackiwTomboulis effective action GammaS for composite operators S is the full fermion propagator, the phase structure of the massless 2 1dimensional Thirring model with fourcomponent spinors is investigated in the HartreeFock HF approximation. In this case both GammaS and its stationary or HF equation for the full fermion propagator S are calculated in the first order of the bare coupling constant G. We have shown that there exist a welldefined dependence of Gequiv GLambda on the cutoff parameter Lambda under which the HF equation is renormalized. In general, it has two sets, i and ii, of solutions for fermion propagator corresponding to dynamical appearance of different mass terms in the model. In the case of set i the mass terms are Hermitian, but the solutions from the set ii correspond to a dynamical generation of the nonHermitiam mass terms, i.e. to a spontaneous nonHermiticity of the Thirring model. Despite this, the mass spectrum of the quasiparticle excitations of all nonHermitian ground states is real. In addition, among these nonHermitian phases there are both cPcT symmetrical and nonsymmetrical phases. Moreover, in contrast with previous investigations of this effect in other models, we have observed the spontaneous nonHermiticity phenomenon also in the it massive 21dimensional Thirring model.

Denoising after Entropybased Debiasing A Robust Training Method for Dataset Bias with Noisy Labels ; Improperly constructed datasets can result in inaccurate inferences. For instance, models trained on biased datasets perform poorly in terms of generalization i.e., dataset bias. Recent debiasing techniques have successfully achieved generalization performance by underestimating easytolearn samples i.e., biasaligned samples and highlighting difficulttolearn samples i.e., biasconflicting samples. However, these techniques may fail owing to noisy labels, because the trained model recognizes noisy labels as difficulttolearn and thus highlights them. In this study, we find that earlier approaches that used the provided labels to quantify difficulty could be affected by the small proportion of noisy labels. Furthermore, we find that running denoising algorithms before debiasing is ineffective because denoising algorithms reduce the impact of difficulttolearn samples, including valuable biasconflicting samples. Therefore, we propose an approach called denoising after entropybased debiasing, i.e., DENEB, which has three main stages. 1 The prejudice model is trained by emphasizing biasaligned, clean samples, which are selected using a Gaussian Mixture Model. 2 Using the persample entropy from the output of the prejudice model, the sampling probability of each sample that is proportional to the entropy is computed. 3 The final model is trained using existing denoising algorithms with the minibatches constructed by following the computed sampling probability. Compared to existing debiasing and denoising algorithms, our method achieves better debiasing performance on multiple benchmarks.

I2MVFormer Large Language Model Generated MultiView Document Supervision for ZeroShot Image Classification ; Recent works have shown that unstructured text documents from online sources can serve as useful auxiliary information for zeroshot image classification. However, these methods require access to a highquality source like Wikipedia and are limited to a single source of information. Large Language Models LLM trained on webscale text show impressive abilities to repurpose their learned knowledge for a multitude of tasks. In this work, we provide a novel perspective on using an LLM to provide text supervision for a zeroshot image classification model. The LLM is provided with a few text descriptions from different annotators as examples. The LLM is conditioned on these examples to generate multiple text descriptions for each classreferred to as views. Our proposed model, I2MVFormer, learns multiview semantic embeddings for zeroshot image classification with these class views. We show that each text view of a class provides complementary information allowing a model to learn a highly discriminative class embedding. Moreover, we show that I2MVFormer is better at consuming the multiview text supervision from LLM compared to baseline models. I2MVFormer establishes a new stateoftheart on three public benchmark datasets for zeroshot image classification with unsupervised semantic embeddings.

Efficient Stein Variational Inference for Reliable Distributionlossless Network Pruning ; Network pruning is a promising way to generate light but accurate models and enable their deployment on resourcelimited edge devices. However, the current stateoftheart assumes that the effective subnetwork and the other superfluous parameters in the given network share the same distribution, where pruning inevitably involves a distribution truncation operation. They usually eliminate values near zero. While simple, it may not be the most appropriate method, as effective models may naturally have many small values associated with them. Removing nearzero values already embedded in model space may significantly reduce model accuracy. Another line of work has proposed to assign discrete prior over all possible substructures that still rely on humancrafted prior hypotheses. Worse still, existing methods use regularized point estimates, namely Hard Pruning, that can not provide error estimations and fail reliability justification for the pruned networks. In this paper, we propose a novel distributionlossless pruning method, named DLLP, to theoretically find the pruned lottery within Bayesian treatment. Specifically, DLLP remodels the vanilla networks as discrete priors for the latent pruned model and the other redundancy. More importantly, DLLP uses Stein Variational Inference to approach the latent prior and effectively bypasses calculating KL divergence with unknown distribution. Extensive experiments based on small Cifar10 and largescaled ImageNet demonstrate that our method can obtain sparser networks with great generalization performance while providing quantified reliability for the pruned model.

fT cosmology against the cosmographic method A new study using mock and observational data ; In this paper, we study the powerlaw fT model using Hubble diagrams of type Ia supernovae SNIa, quasars QSOs, Gamma Ray Bursts GRBs and the measurements from baryonic acoustic oscillations BAO in the framework of the cosmographic method. Using mock data for SNIa, QSOs and GRBs generated based on the powerlaw fT model, we show whether different cosmographic methods are suitable to reconstruct the distance modulus or not. In particular, we investigate the rational PADE polynomials 3,2 and 2,2 in addition to the fourth and fifth order Taylor series. We show that PADE 3,2 is the best approximation that can be used in the cosmographic method to reconstruct the distance modulus at both low and high redshifts. In the context of PADE 3,2 cosmographic method, we show that the powerlaw fT model is well consistent with the real observational data from the Hubble diagrams of SNIa, QSOs and GRBs. Moreover, we find that the combination of the Hubble diagram of SNIa and the BAO observation leads to better consistency between the modelindependent cosmographic method and the powerlaw fT model. Finally, our observational constraints on the parameter of the effective equation of state of DE, described by the powerlaw fT model, show the phantomlike behavior, especially when the BAO observations are included in our analysis.

Genie Show Me the Data for Quantization ; Zeroshot quantization is a promising approach for developing lightweight deep neural networks when data is inaccessible owing to various reasons, including cost and issues related to privacy. By exploiting the learned parameters mu and sigma of batch normalization layers in an FP32pretrained model, zeroshot quantization schemes focus on generating synthetic data. Subsequently, they distill knowledge from the pretrained model teacher to the quantized model student such that the quantized model can be optimized with the synthetic dataset. However, thus far, zeroshot quantization has primarily been discussed in the context of quantizationaware training methods, which require taskspecific losses and longterm optimization as much as retraining. We thus introduce a posttraining quantization scheme for zeroshot quantization that produces highquality quantized networks within a few hours. Furthermore, we propose a framework called Geniethat generates data suited for quantization. With the data synthesized by Genie, we can produce robust quantized models without real datasets, which is comparable to fewshot quantization. We also propose a posttraining quantization algorithm to enhance the performance of quantized models. By combining them, we can bridge the gap between zeroshot and fewshot quantization while significantly improving the quantization performance compared to that of existing approaches. In other words, we can obtain a unique stateoftheart zeroshot quantization approach. The code is available at urlhttpsgithub.comSamsungLabsGenie.

PsychoLinguistic Features Meet Transformer Models for Improved Explainable and Controllable Text Simplification ; Stateoftheart text simplification TS systems adopt endtoend neural network models to directly generate the simplified version of the input text, and usually function as a blackbox. Moreover, TS is usually treated as an allpurpose generic task under the assumption of homogeneity, where the same simplification is suitable for all. In recent years, however, there has been increasing recognition of the need to adapt the simplification techniques to the specific needs of different target groups. In this work, we aim to advance current research on explainable and controllable TS in two ways First, building on recently proposed work to increase the transparency of TS systems, we use a large set of psycholinguistic features in combination with pretrained language models to improve explainable complexity prediction. Second, based on the results of this preliminary task, we extend a stateoftheart Seq2Seq TS model, ACCESS, to enable explicit control of ten attributes. The results of experiments show 1 that our approach improves the performance of stateoftheart models for predicting explainable complexity and 2 that explicitly conditioning the Seq2Seq model on ten attributes leads to a significant improvement in performance in both withindomain and outofdomain settings.

Explicit derivation of the chiral and generic helical edge states for the KaneMele model Closed expressions for the wave function, dispersion relation, and spin rotation ; While one of the most important and intriguing features of the topological insulators is the presence of edge states, the closedform expressions for the edge states of some famous topological models are still lacking. Here, we focus on the KaneMele model with and without Rashba spinorbit coupling as a wellknown model to describe a twodimensional version of the mathbbZ2 topological insulator to study the properties of its edge states analytically. By considering the tightbinding model on a honeycomb lattice with zigzag boundaries and introducing a perturbative approach, we derive explicit expressions for the wave functions, energy dispersion relations, and the spin rotations of the generic helical edge states. To this end, we first map the edge states of the ribbon geometry into an effective twoleg ladder model with momentumdependent energy parameters. Then, we split the Hamiltonian of the system into an unperturbed part and a perturbation. The unperturbed part has a flatband energy spectrum and can be solved exactly which allows us to consider the remaining part of the Hamiltonian perturbatively. The resulting energy dispersion relation within the firstorder perturbation, surprisingly, is in excellent agreement with the numerical spectra over a very wide range of wavenumbers. Our perturbative framework also allows deriving an explicit form for the rotation of the spins of the momentum edge states in the absence of axial spin symmetry due to the Rashba spinorbit interaction.

Probabilistic activity driven model of temporal simplicial networks and its application on higherorder dynamics ; Network modeling characterizes the underlying principles of structural properties and is of vital significance for simulating dynamical processes in real world. However, bridging structure and dynamics is always challenging due to the multiple complexities in real systems. Here, through introducing the individual's activity rate and the possibility of group interaction, we propose a probabilistic activity driven PAD model that could generate temporal higherorder networks with both powerlaw and highclustering characteristics, which successfully links the two most critical structural features and a basic dynamical pattern in extensive complex systems. Surprisingly, the powerlaw exponents and the clustering coefficients of the aggregated PAD network could be tuned in a wide range by altering a set of model parameters. We further provide an approximation algorithm to select the proper parameters that can generate networks with given structural properties, the effectiveness of which is verified by fitting various realworld networks. Lastly, we explore the coevolution of PAD model and higherorder contagion dynamics, and analytically derive the critical conditions for phase transition and bistable phenomenon. Our model provides a basic tool to reproduce complex structural properties and to study the widespread higherorder dynamics, which has great potential for applications across fields.

Electroweak baryogenesis in the threeloop neutrino mass model with dark matter ; Baryon asymmetry of the Universe is evaluated in the model originally proposed in Phys. Rev. Lett. 102 2009 051805, where Majorana masses of neutrinos are generated via threeloop diagrams composed of additional scalar bosons including the dark matter candidate which is odd under an unbroken Z2 symmetry. In order for the model to include multiple CPviolating phases, we do not impose the softly broken Z2 symmetry imposed in the original model to avoid the flavorchanging neutral current at tree level. Instead, for simplicity, we assume the flavor alignment structure in the Yukawa interactions. We also simply assume the alignment structure in the Higgs potential so that the Higgs couplings coincide with those in the SM at tree level. Under these phenomenological simplifications, the model still contains multiple CPviolating phases. By using destructive interferences among them, it is compatible with the stringent constraint from the electric dipole moment measurements to generate the observed baryon asymmetry along with the scenario of electroweak baryogenesis. We show a benchmark scenario which can explain neutrino mass, dark matter and baryon asymmetry of the universe simultaneously and can satisfy all the other available experimental data. Some phenomenological predictions of the model are also discussed.

Addressing Six Standard Model Problems with Technically Natural Higgs Models ; We investigate how many problems of particle physics and cosmology that the recently proposed model framework called Technically Natural Higgs TNH can remedy. In this paper, we will endeavour to answer the following six open questions the electroweak EW naturalness problem, the generation of neutrino masses and their flavour mixing, the nature of the inflaton, the matterantimatter asymmetry problem, the origin of dark matter, and the strong CP problem. In this work, we will consider various possible solutions to these problems for three inflation scenarios Higgs, Starobinsky and scaleindependent inflation in the minimal TNH model, where the Higgs is a mixture of an elementary and a composite state with a compositeness scale much larger than the EW scale. Traditionally, this requires an unnatural small vacuum misalignment, but recently TNH models with compositeness scale up to the Planck scale assisted by a novel mechanism have been proposed. This mechanism is based on softly breaking a global mathbbZ2 symmetry by technically natural small vacuum misalignment, dynamically triggering the EW symmetry breaking and fermion mass generation. With this mechanism, we will show that a scaleinvariant version of the minimal TNH model with a compositeness scale of mathcalO1012 GeV can both provide a technically natural 125 GeV Higgs boson, scotogenic neutrinos, scaleinvariant inflation and QCD axion dark matter, which altogether dynamically induces the Planck scale and may answer all the six abovementioned open questions.

MagicAngle Twisted Symmetric Trilayer Graphene as Topological Heavy Fermion Problem ; Recently, Ref. 1 reformulated magicangle twisted bilayer graphene MATBG as a topological heavy fermion problem, and used this reformulation to provide a deeper understanding for the correlated phases at integer fillings. In this work, we generalize this heavyfermion paradigm to magicangle twisted symmetric trilayer graphene MATSTG, and propose a lowenergy fcd model that reformulates MATSTG as heavy localized f modes coupled to itinerant topological semimetalic c modes and itinerant Dirac d modes. Our fcd model well reproduces the singleparticle band structure of MATSTG at low energies for displacement field mathcalEin0,300meV. By performing HartreeFock calculations with the fcd model for nu0,1,2 electrons per Moir'e unit cell, we reproduce all the correlated ground states obtain from the previous numerical HartreeFock calculations with the BistritzerMacDonaldtype BMtype model, and we find additional new correlated ground states at high displacement field. Based on the numerical results, we propose a simple rule for the ground states at high displacement fields by using the fcd model, and provide analytical derivation for the rule at charge neutrality. We also provide analytical symmetry arguments for the nearlydegenerate energies of the highmathcalE ground states at all the integer fillings of interest, and make experimental predictions of which chargeneutral states are stabilized in magnetic fields. Our fcd model provides a new perspective for understanding the correlated phenomena in MATSTG, suggesting that the heavy fermion paradigm of Ref. 1 should be the generic underpinning of correlated physics in multilayer moire graphene structures.

A Generalized Estimating Equation Approach to Network Regression ; Regression models applied to network data where node attributes are the dependent variables poses a methodological challenge. As has been well studied, naive regression neither properly accounts for community structure, nor does it account for the dependent variable acting as both model outcome and covariate. To address this methodological gap, we propose a network regression model motivated by the important observation that controlling for community structure can, when a network is modular, significantly account for meaningful correlation between observations induced by network connections. We propose a generalized estimating equation GEE approach to learn model parameters based on clusters defined through any singlemembership community detection algorithm applied to the observed network. We provide a necessary condition on the network size and edge formation probabilities to establish the asymptotic normality of the model parameters under the assumption that the graph structure is a stochastic block model. We evaluate the performance of our approach through simulations and apply it to estimate the joint impact of baseline covariates and network effects on COVID19 incidence rate among countries connected by a network of commercial airline traffic. We find that during the beginning of the pandemic the network effect has some influence, the percentage of urban population has more influence on the incidence rate compared to the network effect after the travel ban was in effect.

Amplitude expansion of the phasefield crystal model for complex crystal structures ; The phasefield crystal PFC model describes crystal lattices at diffusive timescales. Its amplitude expansion APFC can be applied to the investigation of relatively large systems under some approximations. However, crystal symmetries accessible within the APFC model are limited to basic ones, namely triangular and square in two dimensions, and bodycentered cubic and facecentered cubic in three dimensions. In this work, we propose a general, amplitudesbased description of virtually any lattice symmetry. To fully exploit the advantages of this model, featuring slowly varying quantities in bulk and localized significant variations at dislocations and interfaces, we consider formulations suitable for realspace numerical methods supporting adaptive spatial discretization. We explore approaches originally proposed for the PFC model which allow for symmetries beyond basic ones through extended parametrizations. Moreover, we tackle the modeling of nonBravais lattices by introducing an amplitude expansion for lattices with a basis and further generalizations. We study and discuss the stability of selected, prototypical lattice symmetries. As pivotal examples, we show that the proposed approach allows for a coarsegrained description of the kagome lattice, exotic square arrangements, and the diamond lattice, as bulk crystals and, importantly, hosting dislocations.

Jamming Attacks on Decentralized Federated Learning in General MultiHop Wireless Networks ; Decentralized federated learning DFL is an effective approach to train a deep learning model at multiple nodes over a multihop network, without the need of a server having direct connections to all nodes. In general, as long as nodes are connected potentially via multiple hops, the DFL process will eventually allow each node to experience the effects of models from all other nodes via either direct connections or multihop paths, and thus is able to train a highfidelity model at each node. We consider an effective attack that uses jammers to prevent the model exchanges between nodes. There are two attack scenarios. First, the adversary can attack any link under a certain budget. Once attacked, two end nodes of a link cannot exchange their models. Secondly, some jammers with limited jamming ranges are deployed in the network and a jammer can only jam nodes within its jamming range. Once a directional link is attacked, the receiver node cannot receive the model from the transmitter node. We design algorithms to select links to be attacked for both scenarios. For the second scenario, we also design algorithms to deploy jammers at optimal locations so that they can attack critical nodes and achieve the highest impact on the DFL process. We evaluate these algorithms by using wireless signal classification over a large network area as the use case and identify how these attack mechanisms exploits various learning, connectivity, and sensing aspects. We show that the DFL performance can be significantly reduced by jamming attacks launched in a wireless network and characterize the attack surface as a vulnerability study before the safe deployment of DFL over wireless networks.

Remnant black hole properties from numericalrelativityinformed perturbation theory and implications for waveform modelling ; During binary black hole BBH mergers, energy and momenta are carried away from the binary system as gravitational radiation. Access to the radiated energy and momenta allows us to predict the properties of the remnant black hole. We develop a Python package gwremnant to compute the remnant mass, remnant spin, peak luminosity, and the final kick imparted on the remnant black hole from the gravitational radiation. Using this package, we compute the remnant properties of the final black hole in case of nonspinning BBH mergers with mass ratios ranging from q2.5 to q1000 using waveform modes generated from BHPTNRSur1dq1e4, a recently developed numericalrelativityinformed surrogate model based on blackhole perturbation theory framework. We validate our results against the remnant properties estimated from numerical relativity NR surrogate models in the comparable mass ratio regime and against recently available highmassratio NR simulations at q15,32,64. We find that our remnant property estimates computed from fluxes at future null infinity closely match the estimates obtained from the NR surrogate model of apparent horizon data. Using Gaussian process regression fitting methods, we train a surrogate model, BHPTNRRemnant, for the properties of the remnant black hole arising from BBH mergers with mass ratios from q2.5 to q1000. Finally, we discuss potential improvements in the BHPTNRSur1dq1e4 waveform model when including remnant information. We make both the gwremnant and BHPTNRRemnant packages publicly available.

Approximate Hofstadter and KapitMuellerlike parent Hamiltonians for Laughlin states on fractals ; Recently, it was shown that fractional quantum Hall states can be defined on fractal lattices. Proposed exact parent Hamiltonians for these states are nonlocal and contain threesite terms. In this work, we look for simpler, approximate parent Hamiltonians for bosonic Laughlin states at half filling, which contain only onsite potentials and twosite hopping with the interaction generated implicitly by hardcore constraints as in the Hofstadter and KapitMueller models on periodic lattices. We use an inverse method'' to determine such Hamiltonians on finitegeneration Sierpi'nski carpet and triangle lattices. The ground states of some of the resulting models display relatively high overlap with the model states if up to third neighbor hopping terms are considered, and by increasing the maximum hopping distance one can achieve nearly perfect overlaps. When the number of particles is reduced and additional potentials are introduced to trap quasiholes, the overlap with a model quasihole wavefunction is also high in some cases, especially for the nonlocal Hamiltonians. We also study how the small system size affects the braiding properties for the model quasihole wavefunctions and perform analogous computations for Hamiltonian models.

Rigid Body Flows for Sampling Molecular Crystal Structures ; Normalizing flows NF are a class of powerful generative models that have gained popularity in recent years due to their ability to model complex distributions with high flexibility and expressiveness. In this work, we introduce a new type of normalizing flow that is tailored for modeling positions and orientations of multiple objects in threedimensional space, such as molecules in a crystal. Our approach is based on two key ideas first, we define smooth and expressive flows on the group of unit quaternions, which allows us to capture the continuous rotational motion of rigid bodies; second, we use the double cover property of unit quaternions to define a proper density on the rotation group. This ensures that our model can be trained using standard likelihoodbased methods or variational inference with respect to a thermodynamic target density. We evaluate the method by training Boltzmann generators for two molecular examples, namely the multimodal density of a tetrahedral system in an external field and the ice XI phase in the TIP4P water model. Our flows can be combined with flows operating on the internal degrees of freedom of molecules and constitute an important step towards the modeling of distributions of many interacting molecules.

Diverse, Difficult, and Odd Instances D2O A New Test Set for Object Classification ; Test sets are an integral part of evaluating models and gauging progress in object recognition, and more broadly in computer vision and AI. Existing test sets for object recognition, however, suffer from shortcomings such as bias towards the ImageNet characteristics and idiosyncrasies e.g., ImageNetV2, being limited to certain types of stimuli e.g., indoor scenes in ObjectNet, and underestimating the model performance e.g., ImageNetA. To mitigate these problems, we introduce a new test set, called D2O, which is sufficiently different from existing test sets. Images are a mix of generated images as well as images crawled from the web. They are diverse, unmodified, and representative of realworld scenarios and cause stateoftheart models to misclassify them with high confidence. To emphasize generalization, our dataset by design does not come paired with a training set. It contains 8,060 images spread across 36 categories, out of which 29 appear in ImageNet. The best Top1 accuracy on our dataset is around 60 which is much lower than 91 best Top1 accuracy on ImageNet. We find that popular vision APIs perform very poorly in detecting objects over D2O categories such as faces'', cars'', and cats''. Our dataset also comes with a miscellaneous'' category, over which we test the image tagging models. Overall, our investigations demonstrate that the D2O test set contain a mix of images with varied levels of difficulty and is predictive of the averagecase performance of models. It can challenge object recognition models for years to come and can spur more research in this fundamental area.

The unreasonable effectiveness of fewshot learning for machine translation ; We demonstrate the potential of fewshot translation systems, trained with unpaired language data, for both high and lowresource language pairs. We show that with only 5 examples of highquality translation data shown at inference, a transformer decoderonly model trained solely with selfsupervised learning, is able to match specialized supervised stateoftheart models as well as more general commercial translation systems. In particular, we outperform the best performing system on the WMT'21 English Chinese news translation task by only using five examples of English Chinese parallel data at inference. Moreover, our approach in building these models does not necessitate joint multilingual training or backtranslation, is conceptually simple and shows the potential to extend to the multilingual setting. Furthermore, the resulting models are two orders of magnitude smaller than stateoftheart language models. We then analyze the factors which impact the performance of fewshot translation systems, and highlight that the quality of the fewshot demonstrations heavily determines the quality of the translations generated by our models. Finally, we show that the fewshot paradigm also provides a way to control certain attributes of the translation we show that we are able to control for regional varieties and formality using only a five examples at inference, paving the way towards controllable machine translation systems.

Flexible, ModelAgnostic Method for Materials Data Extraction from Text Using General Purpose Language Models ; Accurate and comprehensive material databases extracted from research papers are critical for materials science and engineering but require significant human effort to develop. In this paper we present a simple method of extracting materials data from full texts of research papers suitable for quickly developing modestsized databases. The method requires minimal to no coding, prior knowledge about the extracted property, or model training, and provides high recall and almost perfect precision in the resultant database. The method is fully automated except for one humanassisted step, which typically requires just a few hours of human labor. The method builds on top of natural language processing and large general language models but can work with almost any such model. The language models GPT33.5, bart and DeBERTaV3 are evaluated here for comparison. We provide a detailed detailed analysis of the methods performance in extracting bulk modulus data, obtaining up to 90 precision at 96 recall, depending on the amount of human effort involved. We then demonstrate the methods broader effectiveness by developing a database of critical cooling rates for metallic glasses.

CoMAE Single Model Hybrid Pretraining on SmallScale RGBD Datasets ; Current RGBD scene recognition approaches often train two standalone backbones for RGB and depth modalities with the same Places or ImageNet pretraining. However, the pretrained depth network is still biased by RGBbased models which may result in a suboptimal solution. In this paper, we present a singlemodel selfsupervised hybrid pretraining framework for RGB and depth modalities, termed as CoMAE. Our CoMAE presents a curriculum learning strategy to unify the two popular selfsupervised representation learning algorithms contrastive learning and masked image modeling. Specifically, we first build a patchlevel alignment task to pretrain a single encoder shared by two modalities via crossmodal contrastive learning. Then, the pretrained contrastive encoder is passed to a multimodal masked autoencoder to capture the finer context features from a generative perspective. In addition, our singlemodel design without requirement of fusion module is very flexible and robust to generalize to unimodal scenario in both training and testing phases. Extensive experiments on SUN RGBD and NYUDv2 datasets demonstrate the effectiveness of our CoMAE for RGB and depth representation learning. In addition, our experiment results reveal that CoMAE is a dataefficient representation learner. Although we only use the smallscale and unlabeled training set for pretraining, our CoMAE pretrained models are still competitive to the stateoftheart methods with extra largescale and supervised RGB dataset pretraining. Code will be released at httpsgithub.comMCGNJUCoMAE.

Incorporating Expert Opinion on Observable Quantities into Statistical Models A General Framework ; This article describes an approach to incorporate expert opinion on observable quantities through the use of a loss function which updates a prior belief as opposed to specifying parameters on the priors. Eliciting information on observable quantities allows experts to provide meaningful information on a quantity familiar to them, in contrast to elicitation on model parameters, which may be subject to interactions with other parameters or nonlinear transformations before obtaining an observable quantity. The approach to incorporating expert opinion described in this paper is distinctive in that we do not specify a prior to match an expert's opinion on observed quantity, rather we obtain a posterior by updating the model parameters through a loss function. This loss function contains the observable quantity, expressed a function of the parameters, and is related to the expert's opinion which is typically operationalized as a statistical distribution. Parameters which generate observable quantities which are further from the expert's opinion incur a higher loss, allowing for the model parameters to be estimated based on their fidelity to both the data and expert opinion, with the relative strength determined by the number of observations and precision of the elicited belief. Including expert opinion in this fashion allows for a flexible specification of the opinion and in many situations is straightforward to implement with commonly used probabilistic programming software. We highlight this using three worked examples of varying model complexity including survival models, a multivariate normal distribution and a regression problem.