Datasets:

sswt

/

arxiv_sample_50K

like 0

LowLatency Incremental TexttoSpeech Synthesis with Distilled Context Prediction Network ; Incremental texttospeech TTS synthesis generates utterances in small linguistic units for the sake of realtime and lowlatency applications. We previously proposed an incremental TTS method that leverages a large pretrained language model to take unobserved future context into account without waiting for the subsequent segment. Although this method achieves comparable speech quality to that of a method that waits for the future context, it entails a huge amount of processing for sampling from the language model at each time step. In this paper, we propose an incremental TTS method that directly predicts the unobserved future context with a lightweight model, instead of sampling words from the largescale language model. We perform knowledge distillation from a GPT2based context prediction network into a simple recurrent model by minimizing a teacherstudent loss defined between the context embedding vectors of those models. Experimental results show that the proposed method requires about ten times less inference time to achieve comparable synthetic speech quality to that of our previous method, and it can perform incremental synthesis much faster than the average speaking speed of human English speakers, demonstrating the availability of our method to realtime applications.

Energy transport in Z3 chiral clock model ; We characterize the energy transport in a one dimensional mathbbZ3 chiral clock model. The model generalizes the mathbbZ2 symmetric transverse field Ising model TFIM. The model is parametrized by a chirality parameter theta, in addition to f and J which are analogous to the transverse field and the nearest neighbour spin coupling in the TFIM. Unlike the well studied TFIM and XYZ models, does not transform to a fermionic system. We use a matrix product states implementation of the Lindblad master equation to obtain the nonequilibrium steady state NESS in systems of sizes up to 48. We present the estimated NESS current and its scaling exponent gamma as a function of theta at different fJ. The estimated gammafJ,theta point to a ballistic energy transport along a line of integrable points fJcos3theta in the parameter space; all other points deviate from ballistic transport. Analysis of finite size effects within the available system sizes suggest a diffusive behavior away from the integrable points.

MLIMC Machine learningbased implicitsolvent Monte Carlo ; Monte Carlo MC methods are important computational tools for molecular structure optimizations and predictions. When solvent effects are explicitly considered, MC methods become very expensive due to the large degree of freedom associated with the water molecules and mobile ions. Alternatively implicitsolvent MC can largely reduce the computational cost by applying a mean field approximation to solvent effects and meanwhile maintains the atomic detail of the target molecule. The two most popular implicitsolvent models are the PoissonBoltzmann PB model and the Generalized Born GB model in a way such that the GB model is an approximation to the PB model but is much faster in simulation time. In this work, we develop a machine learningbased implicitsolvent Monte Carlo MLIMC method by combining the advantages of both implicit solvent models in accuracy and efficiency. Specifically, the MLIMC method uses a fast and accurate PBbased machine learning PBML scheme to compute the electrostatic solvation free energy at each step. We validate our MLIMC method by using a benzenewater system and a proteinwater system. We show that the proposed MLIMC method has great advantages in speed and accuracy for molecular structure optimization and prediction.

Gaussian ARMA models in the ts.extend package ; This paper introduces and describes the R package ts.extend, which adds probability functions for stationary Gaussian ARMA models and some related utility functions for timeseries. We show how to use the package to compute the density and distributions functions for models in this class, and generate random vectors from this model. The package allows the user to use marginal or conditional models using a simple syntax for conditioning variables and marginalised elements. This allows users to simulate timeseries vectors from any stationary Gaussian ARMA model, even if some elements are conditional values or omitted values. We also show how to use the package to compute the spectral intensity of a timeseries vector and implement the permutationspectrum test for a timeseries vector to detect the presence of a periodic signal.

UQ Accurate Uncertainty Quantification via Anchor Marginalization ; We present DeltaUQ a novel, generalpurpose uncertainty estimator using the concept of anchoring in predictive models. Anchoring works by first transforming the input into a tuple consisting of an anchor point drawn from a prior distribution, and a combination of the input sample with the anchor using a pretext encoding scheme. This encoding is such that the original input can be perfectly recovered from the tuple regardless of the choice of the anchor. Therefore, any predictive model should be able to predict the target response from the tuple alone since it implicitly represents the input. Moreover, by varying the anchors for a fixed sample, we can estimate uncertainty in the prediction even using only a single predictive model. We find this uncertainty is deeply connected to improper sampling of the input data, and inherent noise, enabling us to estimate the total uncertainty in any system. With extensive empirical studies on a variety of usecases, we demonstrate that DeltaUQ outperforms several competitive baselines. Specifically, we study model fitting, sequential model optimization, model based inversion in the regression setting and out of distribution detection, calibration under distribution shifts for classification.

HYPER Learned Hybrid Trajectory Prediction via Factored Inference and Adaptive Sampling ; Modeling multimodal highlevel intent is important for ensuring diversity in trajectory prediction. Existing approaches explore the discrete nature of human intent before predicting continuous trajectories, to improve accuracy and support explainability. However, these approaches often assume the intent to remain fixed over the prediction horizon, which is problematic in practice, especially over longer horizons. To overcome this limitation, we introduce HYPER, a general and expressive hybrid prediction framework that models evolving human intent. By modeling traffic agents as a hybrid discretecontinuous system, our approach is capable of predicting discrete intent changes over time. We learn the probabilistic hybrid model via a maximum likelihood estimation problem and leverage neural proposal distributions to sample adaptively from the exponentially growing discrete space. The overall approach affords a better tradeoff between accuracy and coverage. We train and validate our model on the Argoverse dataset, and demonstrate its effectiveness through comprehensive ablation studies and comparisons with stateoftheart models.

Language Modeling using LMUs 10x Better Data Efficiency or Improved Scaling Compared to Transformers ; Recent studies have demonstrated that the performance of transformers on the task of language modeling obeys a powerlaw relationship with model size over six orders of magnitude. While transformers exhibit impressive scaling, their performance hinges on processing large amounts of data, and their computational and memory requirements grow quadratically with sequence length. Motivated by these considerations, we construct a Legendre Memory Unit based model that introduces a general prior for sequence processing and exhibits an On and On ln n or better dependency for memory and computation respectively. Over three orders of magnitude, we show that our new architecture attains the same accuracy as transformers with 10x fewer tokens. We also show that for the same amount of training our model improves the loss over transformers about as much as transformers improve over LSTMs. Additionally, we demonstrate that adding global selfattention complements our architecture and the augmented model improves performance even further.

Towards Universal Neural Vocoding with a Multiband Excited WaveNet ; This paper introduces the MultiBand Excited WaveNet a neural vocoder for speaking and singing voices. It aims to advance the state of the art towards an universal neural vocoder, which is a model that can generate voice signals from arbitrary mel spectrograms extracted from voice signals. Following the success of the DDSP model and following the development of the recently proposed excitation vocoders we propose a vocoder structure consisting of multiple specialized DNN that are combined with dedicated signal processing components. All components are implemented as differentiable operators and therefore allow joined optimization of the model parameters. To prove the capacity of the model to reproduce high quality voice signals we evaluate the model on single and multi speakersinger datasets. We conduct a subjective evaluation demonstrating that the models support a wide range of domain variations unseen voices, languages, expressivity achieving perceptive quality that compares with a state of the art universal neural vocoder, however using significantly smaller training datasets and significantly less parameters. We also demonstrate remaining limits of the universality of neural vocoders e.g. the creation of saturated singing voices.

Improving Gender Fairness of PreTrained Language Models without Catastrophic Forgetting ; Existing studies addressing gender bias of pretrained language models, usually build a small genderneutral data set and conduct a second phase pretraining on the model with such data. However, given the limited size and concentrated focus of the genderneutral data, catastrophic forgetting would occur during secondphase pretraining. Forgetting information in the original training data may damage the model's downstream performance by a large margin. In this work, we empirically show that catastrophic forgetting occurs in such methods by evaluating them with general NLP tasks in GLUE. Then, we propose a new method, GEnder Equality Prompt GEEP, to improve gender fairness of pretrained models with less forgetting. GEEP freezes the pretrained model and learns genderrelated prompts with genderneutral data. Empirical results show that GEEP not only achieves SOTA performances on gender fairness tasks, but also forgets less and performs better on GLUE by a large margin.

A Modeling Framework for Efficient Reduced Order Simulations of Parametrized LithiumIon Battery Cells ; In this contribution we present a new modeling and simulation framework for parametrized Lithiumion battery cells. We first derive a new continuum model for a rather general intercalation battery cell on the basis of nonequilibrium thermodynamics. In order to efficiently evaluate the resulting parameterized nonlinear system of partial differential equations the reduced basis method is employed. The reduced basis method is a model order reduction technique on the basis of an incremental hierarchical approximate proper orthogonal decomposition approach and empirical operator interpolation. The modeling framework is particularly well suited to investigate and quantify degradation effects of battery cells. Several numerical experiments are given to demonstrate the scope and efficiency of the modeling framework.

Teaching Models new APIs DomainAgnostic Simulators for Task Oriented Dialogue ; We demonstrate that large language models are able to simulate Task Oriented Dialogues in novel domains, provided only with an API implementation and a list of goals. We show these simulations can formulate online, automatic metrics that correlate well with human evaluations. Furthermore, by checking for whether the User's goals are met, we can use simulation to repeatedly generate training data and improve the quality of simulations themselves. With no human intervention or domainspecific training data, our simulations bootstrap endtoend models which achieve a 37 error reduction in previously unseen domains. By including as few as 32 domainspecific conversations, bootstrapped models can match the performance of a fullysupervised model with 10times more data. To our knowledge, this is the first time simulations have been shown to be effective at bootstrapping models without explicitly requiring any domainspecific training data, ruleengineering, or humansintheloop.

Neural AttentionAware Hierarchical Topic Model ; Neural topic models NTMs apply deep neural networks to topic modelling. Despite their success, NTMs generally ignore two important aspects 1 only documentlevel word count information is utilized for the training, while more finegrained sentencelevel information is ignored, and 2 external semantic knowledge regarding documents, sentences and words are not exploited for the training. To address these issues, we propose a variational autoencoder VAE NTM model that jointly reconstructs the sentence and document word counts using combinations of bagofwords BoW topical embeddings and pretrained semantic embeddings. The pretrained embeddings are first transformed into a common latent topical space to align their semantics with the BoW embeddings. Our model also features hierarchical KL divergence to leverage embeddings of each document to regularize those of their sentences, thereby paying more attention to semantically relevant sentences. Both quantitative and qualitative experiments have shown the efficacy of our model in 1 lowering the reconstruction errors at both the sentence and document levels, and 2 discovering more coherent topics from realworld datasets.

Dynamical nonGaussian modelling of spatial processes ; Spatiotemporal processes in environmental applications are often assumed to follow a Gaussian model, possibly after some transformation. However, heterogeneity in space and time might have a pattern that will not be accommodated by transforming the data. In this scenario, modelling the variance laws is an appealing alternative. This work adds flexibility to the usual Multivariate Dynamic Gaussian model by defining the process as a scale mixture between a Gaussian and logGaussian processes. The scale is represented by a process varying smoothly over space and time which is allowed to depend on covariates. Statespace equations define the dynamics over time for both mean and variance processes resulting infeasible inference and prediction. Analysis of artificial datasets show that the parameters are identifiable and simpler models are well recovered by the general proposed model. The analyses of two important environmental processes, maximum temperature and maximum ozone, illustrate the effectiveness of our proposal in improving the uncertainty quantification in the prediction of spatiotemporal processes.

Fitting threedimensional Laguerre tessellations by hierarchical marked point process models ; We present a general statistical methodology for analysing a Laguerre tessellation data set viewed as a realization of a marked point process model. In the first step, for the points we use a nested sequence of multiscale processes which constitute a flexible parametric class of pairwise interaction point process models. In the second step, for the marksradii conditioned on the points we consider various exponential family models where the canonical sufficient statistic is based on tessellation characteristics. For each step parameter estimation based on maximum pseudolikelihood methods is tractable. Model checking is performed using global envelopes and corresponding tests in the first step and by comparing observed and simulated tessellation characteristics in the second step. We apply our methodology for a 3D Laguerre tessellation data set representing the microstructure of a polycrystalline metallic material, where simulations under a fitted model may substitute expensive laboratory experiments.

An extension of the IsingCurieWeiss model of selforganized criticality with a threshold on the interaction range ; In arXiv1301.6911, Cerf and Gorny constructed a model of selforganized criticality, by introducing an automatic control of the temperature parameter in the generalized Ising CurieWeiss model. The fluctuations of the magnetization of this spin model are of order n34 with a limiting law of the form Cexpx4, as in the critical regime of the CurieWeiss model. In this article, we build upon this model by replacing the meanfield interaction with a onedimensional interaction with a certain range rn which varies as a function of the number n of particles. In the Gaussian case, we show that the selfcritical behaviour observed in the meanfield case extends to interaction ranges rngg n34 and we show that this threshold is sharp, with different fluctuations when the interaction range is of order of n34 or smaller than n34.

Improving Hyperparameter Optimization by Planning Ahead ; Hyperparameter optimization HPO is generally treated as a bilevel optimization problem that involves fitting a probabilistic surrogate model to a set of observed hyperparameter responses, e.g. validation loss, and consequently maximizing an acquisition function using a surrogate model to identify good hyperparameter candidates for evaluation. The choice of a surrogate andor acquisition function can be further improved via knowledge transfer across related tasks. In this paper, we propose a novel transfer learning approach, defined within the context of modelbased reinforcement learning, where we represent the surrogate as an ensemble of probabilistic models that allows trajectory sampling. We further propose a new variant of model predictive control which employs a simple lookahead strategy as a policy that optimizes a sequence of actions, representing hyperparameter candidates to expedite HPO. Our experiments on three metadatasets comparing to stateoftheart HPO algorithms including a modelfree reinforcement learning approach show that the proposed method can outperform all baselines by exploiting a simple planningbased policy.

Breaking Down Multilingual Machine Translation ; While multilingual training is now an essential ingredient in machine translation MT systems, recent work has demonstrated that it has different effects in different multilingual settings, such as manytoone, onetomany, and manytomany learning. These training settings expose the encoder and the decoder in a machine translation model with different data distributions. In this paper, we examine how different varieties of multilingual training contribute to learning these two components of the MT model. Specifically, we compare bilingual models with encoders andor decoders initialized by multilingual training. We show that multilingual training is beneficial to encoders in general, while it only benefits decoders for lowresource languages LRLs. We further find the important attention heads for each language pair and compare their correlations during inference. Our analysis sheds light on how multilingual translation models work and enables us to propose methods to improve performance by training with highly related languages. Our manytoone models for highresource languages and onetomany models for LRL outperform the best results reported by Aharoni et al. 2019

Hey AI, Can You Solve Complex Tasks by Talking to Agents ; Training giant models from scratch for each complex task is resource and datainefficient. To help develop models that can leverage existing systems, we propose a new challenge Learning to solve complex tasks by communicating with existing agents or models in natural language. We design a synthetic benchmark, CommaQA, with three complex reasoning tasks explicit, implicit, numeric designed to be solved by communicating with existing QA agents. For instance, using text and table QA agents to answer questions such as Who had the longest javelin throw from USA. We show that blackbox models struggle to learn this task from scratch accuracy under 50 even with access to each agent's knowledge and gold facts supervision. In contrast, models that learn to communicate with agents outperform blackbox models, reaching scores of 100 when given gold decomposition supervision. However, we show that the challenge of learning to solve complex tasks by communicating with existing agents emphwithout relying on any auxiliary supervision or data still remains highly elusive. We release CommaQA, along with a compositional generalization test split, to advance research in this direction. Dataset and Code available at httpsgithub.comallenaicommaqa.

Efficient and Consistent DataDriven Model Selection for Time Series ; This paper studies the model selection problem in a large class of causal time series models, which includes both the ARMA or ARinfty processes, as well as the GARCH or ARCHinfty, APARCH, ARMAGARCH and many others processes. We first study the asymptotic behavior of the ideal penalty that minimizes the risk induced by a quasilikelihood estimation among a finite family of models containing the true model. Then, we provide general conditions on the penalty term for obtaining the consistency and efficiency properties. We notably prove that consistent model selection criteria outperform classical AIC criterion in terms of efficiency. Finally, we derive from a Bayesian approach the usual BIC criterion, and by keeping all the second order terms of the Laplace approximation, a datadriven criterion denoted KC'. MonteCarlo experiments exhibit the obtained asymptotic results and show that KC' criterion does better than the AIC and BIC ones in terms of consistency and efficiency.

Model Composition Can Multiple Neural Networks Be Combined into a Single Network Using Only Unlabeled Data ; The diversity of deep learning applications, datasets, and neural network architectures necessitates a careful selection of the architecture and data that match best to a target application. As an attempt to mitigate this dilemma, this paper investigates the idea of combining multiple trained neural networks using unlabeled data. In addition, combining multiple models into one can speed up the inference, result in stronger, more capable models, and allows us to select efficient devicefriendly target network architectures. To this end, the proposed method makes use of generation, filtering, and aggregation of reliable pseudolabels collected from unlabeled data. Our method supports using an arbitrary number of input models with arbitrary architectures and categories. Extensive performance evaluations demonstrated that our method is very effective. For example, for the task of object detection and without using any groundtruth labels, an EfficientDetD0 trained on PascalVOC and an EfficientDetD1 trained on COCO, can be combined to a RetinaNetResNet50 model, with a similar mAP as the supervised training. If finetuned in a semisupervised setting, the combined model achieves 18.6, 12.6, and 8.1 mAP improvements over supervised training with 1, 5, and 10 of labels.

Sampling from Arbitrary Functions via PSD Models ; In many areas of applied statistics and machine learning, generating an arbitrary number of independent and identically distributed i.i.d. samples from a given distribution is a key task. When the distribution is known only through evaluations of the density, current methods either scale badly with the dimension or require very involved implementations. Instead, we take a twostep approach by first modeling the probability distribution and then sampling from that model. We use the recently introduced class of positive semidefinite PSD models, which have been shown to be efficient for approximating probability densities. We show that these models can approximate a large class of densities concisely using few evaluations, and present a simple algorithm to effectively sample from these models. We also present preliminary empirical results to illustrate our assertions.

Bayesian Analysis of Stochastic Volatility Model using Finite Gaussian Mixtures with Unknown Number of Components ; Financial studies require volatility based models which provides useful insights on risks related to investments. Stochastic volatility models are one of the most popular approaches to model volatility in such studies. The asset returns under study may come in multiple clusters which are not captured well assuming standard distributions. Mixture distributions are more appropriate in such situations. In this work, an algorithm is demonstrated which is capable of studying finite mixtures but with unknown number of components. This algorithm uses a BirthDeath process to adjust the number of components in the mixture distribution and the weights are assigned accordingly. This mixture distribution specification is then used for asset returns and a semiparametric stochastic volatility model is fitted in a Bayesian framework. A specific case of Gaussian mixtures is studied. Using appropriate prior specification, Gibbs sampling method is used to generate posterior chains and assess model convergence. A case study of stock return data for State Bank of India is used to illustrate the methodology.

Categorizing models using SelfOrganizing Maps an application to modified gravity theories probed by cosmic shear ; We propose to use SelfOrganizing Maps SOM to map the impact of physical models onto observables. Using this approach, we are be able to determine how theories relate to each other given their signatures. In cosmology this will be particularly useful to determine cosmological models such as dark energy, modified gravity or inflationary models that should be tested by the new generation of experiments. As a first example, we apply this approach to the representation of a subset of the space of modified gravity theories probed by cosmic shear. We therefore train a SOM on shear correlation functions in the fR, dilaton and symmetron models. The results indicate these three theories have similar signatures on shear for small values of their parameters but the dilaton has different signature for higher values. We also show that modified gravity especially the dilaton model has a different impact on cosmic shear compared to a dynamical dark energy so both need to be tested by galaxy surveys.

Extracting Expert's Goals by Whatif Interpretable Modeling ; Although reinforcement learning RL has tremendous success in many fields, applying RL to realworld settings such as healthcare is challenging when the reward is hard to specify and no exploration is allowed. In this work, we focus on recovering clinicians' rewards in treating patients. We incorporate the whatif reasoning to explain the clinician's treatments based on their potential future outcomes. We use generalized additive models GAMs a class of accurate, interpretable models to recover the reward. In both simulation and a realworld hospital dataset, we show our model outperforms baselines. Finally, our model's explanations match several clinical guidelines when treating patients while we found the commonlyused linear model often contradicts them.

Phenomenological analysis of multipseudoscalar mediated dark matter models ; Nonminimal simplified extensions of the Standard Model have gained considerable currency in the context of dark matter searches at the LHC, since they predict enhanced monoHiggs and monoWZ signatures over large parts of the parameter space. However, these nonminimal models obviously lack the simplicity and directness of the original simplified models, and are more heavily dependent on the model assumptions. We propose to classify these models generically on the basis of additional mediators and dark matter particles. As an example, we take up a scenario involving multiple pseudoscalar mediators, and a single Dirac dark matter particle, the latter being a popular introduction to ensure ultraviolet completion of theories with multiple pseudoscalar fields. In the chosen scenario, we discuss the viable channels and signatures of relevance at the future runs of the LHC. These are then compared with the minimal simplified scenarios and distinguishing features are pinpointed.

Distilling Relation Embeddings from Pretrained Language Models ; Pretrained language models have been found to capture a surprisingly rich amount of lexical knowledge, ranging from commonsense properties of everyday concepts to detailed factual knowledge about named entities. Among others, this makes it possible to distill highquality word vectors from pretrained language models. However, it is currently unclear to what extent it is possible to distill relation embeddings, i.e. vectors that characterize the relationship between two words. Such relation embeddings are appealing because they can, in principle, encode relational knowledge in a more finegrained way than is possible with knowledge graphs. To obtain relation embeddings from a pretrained language model, we encode word pairs using a manually or automatically generated prompt, and we finetune the language model such that relationally similar word pairs yield similar output vectors. We find that the resulting relation embeddings are highly competitive on analogy unsupervised and relation classification supervised benchmarks, even without any taskspecific finetuning. Source code to reproduce our experimental results and the model checkpoints are available in the following repository httpsgithub.comasahi417relbert

Revisiting joint decoding based multitalker speech recognition with DNN acoustic model ; In typical multitalker speech recognition systems, a neural networkbased acoustic model predicts senone state posteriors for each speaker. These are later used by a singletalker decoder which is applied on each speakerspecific output stream separately. In this work, we argue that such a scheme is suboptimal and propose a principled solution that decodes all speakers jointly. We modify the acoustic model to predict joint state posteriors for all speakers, enabling the network to express uncertainty about the attribution of parts of the speech signal to the speakers. We employ a joint decoder that can make use of this uncertainty together with higherlevel language information. For this, we revisit decoding algorithms used in factorial generative models in early multitalker speech recognition systems. In contrast with these early works, we replace the GMM acoustic model with DNN, which provides greater modeling power and simplifies part of the inference. We demonstrate the advantage of joint decoding in proof of concept experiments on a mixedTIDIGITS dataset.

MultiTask Learning based Convolutional Models with Curriculum Learning for the Anisotropic Reynolds Stress Tensor in Turbulent Duct Flow ; The Reynoldsaveraged NavierStokes RANS equations require accurate modeling of the anisotropic Reynolds stress tensor. Traditional closure models, while sophisticated, often only apply to restricted flow configurations. Researchers have started using machine learning approaches to tackle this problem by developing more general closure models informed by data. In this work we build upon recent convolutional neural network architectures used for turbulence modeling and propose a multitask learningbased fully convolutional neural network that is able to accurately predict the normalized anisotropic Reynolds stress tensor for turbulent duct flows. Furthermore, we also explore the application of curriculum learning to datadriven turbulence modeling.

Interpretable and Explainable Machine Learning for Materials Science and Chemistry ; While the uptake of datadriven approaches for materials science and chemistry is at an exciting, early stage, to realise the true potential of machine learning models for successful scientific discovery, they must have qualities beyond purely predictive power. The predictions and inner workings of models should provide a certain degree of explainability by human experts, permitting the identification of potential model issues or limitations, building trust on model predictions and unveiling unexpected correlations that may lead to scientific insights. In this work, we summarize applications of interpretability and explainability techniques for materials science and chemistry and discuss how these techniques can improve the outcome of scientific studies. We discuss various challenges for interpretable machine learning in materials science and, more broadly, in scientific settings. In particular, we emphasize the risks of inferring causation or reaching generalization by purely interpreting machine learning models and the need of uncertainty estimates for model explanations. Finally, we showcase a number of exciting developments in other fields that could benefit interpretability in material science and chemistry problems.

Cosmological models based on a complex scalar field with a powerlaw potential associated with a polytropic equation of state ; We construct cosmological models based on a complex scalar field with a powerlaw potential VfracKgamma1fracmhbar2gammavarphi2gamma associated with a polytropic equation of state PKrhogamma the potential associated with an isothermal equation of state Prho kB Tm is Vfracm kB Thbar2varphi2 lnm2varphi2rhohbar21 and the potential associated with a logotropic equation of state PAlnrhorhoP is VAlnm2varphi2hbar2rhoP1. We consider a fast oscillation regime of spintessence'' where the equations of the problem can be simplified. We study all possible cases with arbitrary positive and negative values of the polytropic constant and polytropic index. The LambdaCDM model, the Chaplygin gas model and the BoseEinstein condensate model are recovered as particular cases of our study corresponding to a constant potential gamma0, an inverse squarelaw potential gamma1, and a quartic potential gamma2. We also derive the twofluid representation of the Chaplygin gas model.

Equivariant Deep Dynamical Model for Motion Prediction ; Learning representations through deep generative modeling is a powerful approach for dynamical modeling to discover the most simplified and compressed underlying description of the data, to then use it for other tasks such as prediction. Most learning tasks have intrinsic symmetries, i.e., the input transformations leave the output unchanged, or the output undergoes a similar transformation. The learning process is, however, usually uninformed of these symmetries. Therefore, the learned representations for individually transformed inputs may not be meaningfully related. In this paper, we propose an SO3 equivariant deep dynamical model EqDDM for motion prediction that learns a structured representation of the input space in the sense that the embedding varies with symmetry transformations. EqDDM is equipped with equivariant networks to parameterize the statespace emission and transition models. We demonstrate the superior predictive performance of the proposed model on various motion data.

A systematic approach for modeling a nonlocal eddy diffusivity ; This study considers advective and diffusive transport of passive scalar fields by spatiallyvarying incompressible flows. Prior studies have shown that the eddy diffusivities governing the mean field transport in such systems can generally be nonlocal in space and time. While for many flows nonlocal eddy diffusivities are more accurate than commonlyused Boussinesq eddy diffusivities, nonlocal eddy diffusivities are often computationally costprohibitive to obtain and difficult to implement in practice. We develop a systematic and more costeffective approach for modeling nonlocal eddy diffusivities using matched moment inverse MMI operators. These operators are constructed using only a few leadingorder moments of the exact nonlocal eddy diffusivity kernel, which can be easily computed using the inverse macroscopic forcing method IMFM Mani and Park 2021. The resulting reducedorder models for the mean fields that incorporate the modeled eddy diffusivities often improve Boussinesqlimit models since they capture leadingorder nonlocal effects. But more importantly, these models can be expressed as partial differential equations that are readily solvable using existing computational fluid dynamics capabilities rather than as integropartial differential equations.

MachineintheLoop Rewriting for Creative Image Captioning ; Machineintheloop writing aims to enable humans to collaborate with models to complete their writing tasks more effectively. Prior work has found that providing humans a machinewritten draft or sentencelevel continuations has limited success since the generated text tends to deviate from humans' intention. To allow the user to retain control over the content, we train a rewriting model that, when prompted, modifies specified spans of text within the user's original draft to introduce descriptive and figurative elements locally in the text. We evaluate the model on its ability to collaborate with humans on the task of creative image captioning. On a user study through Amazon Mechanical Turk, our model is rated to be more helpful than a baseline infilling language model. In addition, thirdparty evaluation shows that users write more descriptive and figurative captions when collaborating with our model compared to completing the task alone.

Pretrained TransformerBased Approach for Arabic Question Answering A Comparative Study ; Question answeringQA is one of the most challenging yet widely investigated problems in Natural Language Processing NLP. Questionanswering QA systems try to produce answers for given questions. These answers can be generated from unstructured or structured text. Hence, QA is considered an important research area that can be used in evaluating text understanding systems. A large volume of QA studies was devoted to the English language, investigating the most advanced techniques and achieving stateoftheart results. However, research efforts in the Arabic questionanswering progress at a considerably slower pace due to the scarcity of research efforts in Arabic QA and the lack of large benchmark datasets. Recently many pretrained language models provided high performance in many Arabic NLP problems. In this work, we evaluate the stateoftheart pretrained transformers models for Arabic QA using four reading comprehension datasets which are ArabicSQuAD, ARCD, AQAD, and TyDiQAGoldP datasets. We finetuned and compared the performance of the AraBERTv2base model, AraBERTv0.2large model, and AraELECTRA model. In the last, we provide an analysis to understand and interpret the lowperformance results obtained by some models.

Going... going... wrong a test of the levelk and cognitive hierarchy models of bidding behaviour ; In this paper, we design and implement an experiment aimed at testing the levelk model of auctions. We begin by asking which simple environments can best disentangle the levelk model from its leading rival, BayesNash equilibrium. We find two environments that are particularly suited to this purpose an allpay auction with uniformly distributed values, and a firstprice auction with the possibility of cancelled bids. We then implement both of these environments in a virtual laboratory in order to see which theory can best explain observed bidding behaviour. We find that, when plausibly calibrated, the levelk model substantially underpredicts the observed bids and is clearly outperformed by equilibrium. Moreover, attempting to fit the levelk model to the observed data results in implausibly high estimated levels, which in turn bear no relation to the levels inferred from a game known to trigger levelk reasoning. Finally, subjects almost never appeal to iterated reasoning when asked to explain how they bid. Overall, these findings suggest that, despite its notable success in predicting behaviour in other strategic settings, the levelk model and its close cousin cognitive hierarchy cannot explain behaviour in auctions.

Analysis of 5G academic Network based on graph representation learning method ; With the rapid development of 5th Generation Mobile Communication Technology 5G, the diverse forms of collaboration and extensive data in academic social networks constructed by 5G papers make the management and analysis of academic social networks increasingly challenging. Despite the particular success achieved by representation learning in analyzing academic and social networks, most present presentation learning models focus on maintaining the firstorder and secondorder similarity of nodes. They rarely possess similar structural characteristics of spatial independence in the network. This paper proposes a Loworder Network representation Learning Model LNLM based on Nonnegative Matrix Factorization NMF to solve these problems. The model uses the random walk method to extract loworder features of nodes and map multiple components to a lowdimensional space, effectively maintaining the internal correlation between members. This paper verifies the performance of this model, conducts comparative experiments on four test datasets and four real network datasets through downstream tasks such as multilabel classification, clustering, and link prediction. Comparing eight mainstream network representation learning models shows that the proposed model can significantly improve the detection efficiency and learning methods and effectively extract local and loworder features of the network.

Machine LearningAssisted Analysis of Small Angle Xray Scattering ; Small angle Xray scattering SAXS is extensively used in materials science as a way of examining nanostructures. The analysis of experimental SAXS data involves mapping a rather simple data format to a vast amount of structural models. Despite various scientific computing tools to assist the model selection, the activity heavily relies on the SAXS analysts' experience, which is recognized as an efficiency bottleneck by the community. To cope with this decisionmaking problem, we develop and evaluate the opensource, Machine Learningbased tool SCAN SCattering Ai aNalysis to provide recommendations on model selection. SCAN exploits multiple machine learning algorithms and uses models and a simulation tool implemented in the SasView package for generating a well defined set of datasets. Our evaluation shows that SCAN delivers an overall accuracy of 9597. The XGBoost Classifier has been identified as the most accurate method with a good balance between accuracy and training time. With eleven predefined structural models for common nanostructures and an easy drawdrop function to expand the number and types training models, SCAN can accelerate the SAXS data analysis workflow.

IVGNN Interval Valued Data Handling Using Graph Neural Network ; Graph Neural Network GNN is a powerful tool to perform standard machine learning on graphs. To have a Euclidean representation of every node in the NonEuclidean graphlike data, GNN follows neighbourhood aggregation and combination of information recursively along the edges of the graph. Despite having many GNN variants in the literature, no model can deal with graphs having nodes with intervalvalued features. This article proposes an IntervalValuedGraph Neural Network, a novel GNN model where, for the first time, we relax the restriction of the feature space being countable. Our model is much more general than existing models as any countable set is always a subset of the universal set Rn, which is uncountable. Here, to deal with intervalvalued feature vectors, we propose a new aggregation scheme of intervals and show its expressive power to capture different interval structures. We validate our theoretical findings about our model for graph classification tasks by comparing its performance with those of the stateoftheart models on several benchmark network and synthetic datasets.

Personalized Federated Learning through Local Memorization ; Federated learning allows clients to collaboratively learn statistical models while keeping their data local. Federated learning was originally used to train a unique global model to be served to all clients, but this approach might be suboptimal when clients' local data distributions are heterogeneous. In order to tackle this limitation, recent personalized federated learning methods train a separate model for each client while still leveraging the knowledge available at other clients. In this work, we exploit the ability of deep neural networks to extract high quality vectorial representations embeddings from nontabular data, e.g., images and text, to propose a personalization mechanism based on local memorization. Personalization is obtained by interpolating a collectively trained global model with a local knearest neighbors kNN model based on the shared representation provided by the global model. We provide generalization bounds for the proposed approach in the case of binary classification, and we show on a suite of federated datasets that this approach achieves significantly higher accuracy and fairness than stateoftheart methods.

Link Cascades in Complex Networks A Meanfield Approach ; Cascade models on networks have been used extensively to study cascade failure in complex systems. However, most current models consider failure caused by node damage and neglect the possibility of link damage, which is relevant to transportation, social dynamics, biology, and medicine. In an attempt to generalize conventional cascade models to link damage, we propose a link cascade model based on the standard independent cascade model, which is then solved via both numerical simulation and analytic approximation. We find that the probability that a node loses all its links due to link damage exhibits a minimum as a function of node degree, indicating that there exists an optimal degree for a node to be most resistant to link damage. We apply our model to investigate the sign distribution in a realworld signed social network and find that such optimal degree does exist in realworld dataset.

Graph Neural Networks with Parallel Neighborhood Aggregations for Graph Classification ; We focus on graph classification using a graph neural network GNN model that precomputes the node features using a bank of neighborhood aggregation graph operators arranged in parallel. These GNN models have a natural advantage of reduced training and inference time due to the precomputations but are also fundamentally different from popular GNN variants that update node features through a sequential neighborhood aggregation procedure during training. We provide theoretical conditions under which a generic GNN model with parallel neighborhood aggregations PAGNNs, in short are provably as powerful as the wellknown WeisfeilerLehman WL graph isomorphism test in discriminating nonisomorphic graphs. Although PAGNN models do not have an apparent relationship with the WL test, we show that the graph embeddings obtained from these two methods are injectively related. We then propose a specialized PAGNN model, called SPIN, which obeys the developed conditions. We demonstrate via numerical experiments that the developed model achieves stateoftheart performance on many diverse realworld datasets while maintaining the discriminative power of the WL test and the computational advantage of preprocessing graphs before the training process.

Geometrically reduced modelling of pulsatile flow in perivascular networks ; Flow of cerebrospinal fluid in perivascular spaces is a key mechanism underlying brain transport and clearance. In this paper, we present a mathematical and numerical formalism for reduced models of pulsatile viscous fluid flow in networks of generalized annular cylinders. We apply this framework to study cerebrospinal fluid flow in perivascular spaces induced by pressure differences, cardiac pulse waveinduced vascular wall motion and vasomotion. The reduced models provide approximations of the crosssection average pressure and crosssection flux, both defined over the topologically onedimensional centerlines of the network geometry. Comparing the full and reduced model predictions, we find that the reduced models capture pulsatile flow characteristics and provide accurate pressure and flux predictions across the range of idealized and imagebased scenarios investigated at a fraction of the computational cost of the corresponding full models. The framework presented thus provides a robust and effective computational approach for large scale insilico studies of pulsatile perivascular fluid flow and transport.

Two models unifying warm inflation with dark matter and dark energy ; Two models that unify warm inflation with dark matter and dark energy are proposed. In the models, a single scalar field is responsible for the early expansion of the universe through the process of dissipative warm inflation and then acts as both dark matter and dark energy in subsequent stages. The first model is based on a noncanonical field with the Lagrangian density mathcalLFXVphi, where the potential is dominant at the slowroll inflationary epoch and negligible in subsequent stages. The second model takes advantage of a kessence Lagrangian density having the coupled form mathcalLFXVphi. For both models, equations of the evolution for the fields and observational constraints are presented, and an evolution law describing how the energy density rho and state parameter w scale with the scale factor a is obtained.

Maximum likelihood estimation for a stochastic SEIR system for COVID19 ; The parameter estimation of epidemic datadriven models is a crucial task. In some cases, we can formulate a better model by describing uncertainty with appropriate noise terms. However, because of the limited extent and partial information, in general this kind of model leads to intractable likelihoods. Here, we illustrate how a stochastic extension of the SEIR model improves the uncertainty quantification of an overestimated MCMC scheme based on its deterministic model to count reportedconfirmed COVID19 cases in Mexico City. Using a particular mechanism to manage missing data, we developed MLE for some parameters of the stochastic model, which improves the description of variance of the actual data.

Explore the Potential Performance of VisionandLanguage Navigation Model a Snapshot Ensemble Method ; VisionandLanguage Navigation VLN is a challenging task in the field of artificial intelligence. Although massive progress has been made in this task over the past few years attributed to breakthroughs in deep vision and language models, it remains tough to build VLN models that can generalize as well as humans. In this paper, we provide a new perspective to improve VLN models. Based on our discovery that snapshots of the same VLN model behave significantly differently even when their success rates are relatively the same, we propose a snapshotbased ensemble solution that leverages predictions among multiple snapshots. Constructed on the snapshots of the existing stateoftheart SOTA model circlearrowrightBERT and our pastactionaware modification, our proposed ensemble achieves the new SOTA performance in the R2R dataset challenge in Navigation Error NE and Success weighted by Path Length SPL.

Gluon string breaking and meson spectrum in the holographic Soft Wall model ; We propose a general method for finding a stringlike meson spectrum which is based on a certain condition for the breaking of closed gluon string. A model is constructed that successfully realizes the proposed method. The model is based on the holographic Soft Wall model for QCD and the use of the Wilson confinement criterion. We applied our approach to the vector and scalar cases and obtained numerical predictions for the intercepts of corresponding Regge like radial meson spectra. A good agreement is obtained both with the existing experimental data and with some other known phenomenological approaches. Remarkably, our closed string breaking condition has two branches. We argue that they should correspond to states of opposite parity. The Wilson confinement criterion leads then to a natural mass splitting between parity partners. The constructed model represents thus the first example of bottomup holographic model in which the effects of chiral symmetry breaking can emerge automatically, i.e., without additional assumptions taken outside of the holographic approach.

Leveraging Intrinsic Gradient Information for Further Training of Differentiable Machine Learning Models ; Designing models that produce accurate predictions is the fundamental objective of machine learning ML. This work presents methods demonstrating that when the derivatives of target variables outputs with respect to inputs can be extracted from processes of interest, e.g., neural networks NN based surrogate models, they can be leveraged to further improve the accuracy of differentiable ML models. This paper generalises the idea and provides practical methodologies that can be used to leverage gradient information GI across a variety of applications including 1 Improving the performance of generative adversarial networks GANs; 2 efficiently tuning NN model complexity; 3 regularising linear regressions. Numerical results show that GI can effective enhance ML models with existing datasets, demonstrating its value for a variety of applications.

Sphere Face ModelA 3D Morphable Model with Hypersphere Manifold Latent Space ; 3D Morphable Models 3DMMs are generative models for face shape and appearance. However, the shape parameters of traditional 3DMMs satisfy the multivariate Gaussian distribution while the identity embeddings satisfy the hypersphere distribution, and this conflict makes it challenging for face reconstruction models to preserve the faithfulness and the shape consistency simultaneously. To address this issue, we propose the Sphere Face ModelSFM, a novel 3DMM for monocular face reconstruction, which can preserve both shape fidelity and identity consistency. The core of our SFM is the basis matrix which can be used to reconstruct 3D face shapes, and the basic matrix is learned by adopting a twostage training approach where 3D and 2D training data are used in the first and second stages, respectively. To resolve the distribution mismatch, we design a novel loss to make the shape parameters have a hyperspherical latent space. Extensive experiments show that SFM has high representation ability and shape parameter space's clustering performance. Moreover, it produces fidelity face shapes, and the shapes are consistent in challenging conditions in monocular face reconstruction.

Hybrid quantumclassical algorithm for computing imaginarytime correlation functions ; Quantitative descriptions of strongly correlated materials pose a considerable challenge in condensed matter physics and chemistry. A promising approach to address this problem is quantum embedding methods. In particular, the dynamical meanfield theory DMFT maps the original system to an effective quantum impurity model comprising correlated orbitals embedded in an electron bath. The biggest bottleneck in DMFT calculations is numerically solving the quantum impurity model, i.e., computing Green's function. Past studies have proposed theoretical methods to compute Green's function of a quantum impurity model in polynomial time using a quantum computer. So far, however, efficient methods for computing the imaginarytime Green's functions have not been established despite the advantages of the imaginarytime formulation. We propose a quantumclassical hybrid algorithm for computing imaginarytime Green's functions on quantum devices with limited hardware resources by applying the variational quantum simulation. Using a quantum circuit simulator, we verified this algorithm by computing Green's functions for a dimer model as well as a foursite impurity model obtained by DMFT calculations of the singleband Hubbard model, although our method can be applied to general imaginarytime correlation functions.

BSM Cosmology from BSM Physics ; Now Standard LambdaCDM cosmology is based on physics Beyond the Standard Model BSM, which in turn needs cosmological probes for its study. This vicious circle of problems can be resolved by methods of cosmoparticle physics, in which cosmological messengers of new physics provide sensitive model dependent probes for BSM physics. Such messengers, which are inevitably present in any BSM basis for now Standard cosmology, lead to deviations from the Standard cosmological paradigm. We give brief review of some possible cosmological features and messengers of BSM physics, which include balancing of baryon asymmetry and dark matter by sphaleron transitions, hadronic dark matter and exotic cosmic ray components, a solution for puzzles of direct dark matter searches in dark atom model, antimatter in baryon asymmetrical Universe as sensitive probe for models of inflation and baryosynthesis and its possible probe in AMS02 experiment, PBH and GW messengers of BSM models and phase transitions in early Universe. These aspects are discussed in the general framework of methods of cosmoparticle physics.

FLAVA A Foundational Language And Vision Alignment Model ; Stateoftheart vision and visionandlanguage models rely on largescale visiolinguistic pretraining for obtaining good performance on a variety of downstream tasks. Generally, such models are often either crossmodal contrastive or multimodal with earlier fusion but not both; and they often only target specific modalities or tasks. A promising direction would be to use a single holistic universal model, as a foundation, that targets all modalities at once a true vision and language foundation model should be good at vision tasks, language tasks, and cross and multimodal vision and language tasks. We introduce FLAVA as such a model and demonstrate impressive performance on a wide range of 35 tasks spanning these target modalities.

Blockwise Sequential Model Learning for Partially Observable Reinforcement Learning ; This paper proposes a new sequential model learning architecture to solve partially observable Markov decision problems. Rather than compressing sequential information at every timestep as in conventional recurrent neural networkbased methods, the proposed architecture generates a latent variable in each data block with a length of multiple timesteps and passes the most relevant information to the next block for policy optimization. The proposed blockwise sequential model is implemented based on selfattention, making the model capable of detailed sequential learning in partial observable settings. The proposed model builds an additional learning network to efficiently implement gradient estimation by using selfnormalized importance sampling, which does not require the complex blockwise input data reconstruction in the model learning. Numerical results show that the proposed method significantly outperforms previous methods in various partially observable environments.

Federated Twostage Learning with Signbased Voting ; Federated learning is a distributed machine learning mechanism where local devices collaboratively train a shared global model under the orchestration of a central server, while keeping all private data decentralized. In the system, model parameters and its updates are transmitted instead of raw data, and thus the communication bottleneck has become a key challenge. Besides, recent larger and deeper machine learning models also pose more difficulties in deploying them in a federated environment. In this paper, we design a federated twostage learning framework that augments prototypical federated learning with a cut layer on devices and uses signbased stochastic gradient descent with the majority vote method on model updates. Cut layer on devices learns informative and lowdimension representations of raw data locally, which helps reduce global model parameters and prevents data leakage. Signbased SGD with the majority vote method for model updates also helps alleviate communication limitations. Empirically, we show that our system is an efficient and privacy preserving federated learning scheme and suits for general application scenarios.

ExModel Continual Learning from a Stream of Trained Models ; Learning continually from nonstationary data streams is a challenging research topic of growing popularity in the last few years. Being able to learn, adapt, and generalize continually in an efficient, effective, and scalable way is fundamental for a sustainable development of Artificial Intelligent systems. However, an agentcentric view of continual learning requires learning directly from raw data, which limits the interaction between independent agents, the efficiency, and the privacy of current approaches. Instead, we argue that continual learning systems should exploit the availability of compressed information in the form of trained models. In this paper, we introduce and formalize a new paradigm named ExModel Continual Learning ExML, where an agent learns from a sequence of previously trained models instead of raw data. We further contribute with three exmodel continual learning algorithms and an empirical setting comprising three datasets MNIST, CIFAR10 and CORe50, and eight scenarios, where the proposed algorithms are extensively tested. Finally, we highlight the peculiarities of the exmodel paradigm and we point out interesting future research directions.

Solving the nonpreemptive two queue polling model with generally distributed service and switchover durations and Poisson arrivals as a SemiMarkov Decision Process ; The polling system with switchover durations is a useful model with several practical applications. It is classified as a Discrete Event Dynamic System DEDS for which no one agreed upon modelling approach exists. Furthermore, DEDS are quite complex. To date, the most sophisticated approach to modelling the polling system of interest has been a Continuoustime Markov Decision Process CTMDP. This paper presents a SemiMarkov Decision Process SMDP formulation of the polling system as to introduce additional modelling power. Such power comes at the expense of truncation errors and expensive numerical integrals which naturally leads to the question of whether the SMDP policy provides a worthwhile advantage. To further add to this scenario, it is shown how sparsity can be exploited in the CTMDP to develop a computationally efficient model. The discounted performance of the SMDP and CTMDP policies are evaluated using a SemiMarkov Process simulator. The two policies are accompanied by a heuristic policy specifically developed for this polling system a well as an exhaustive service policy. Parametric and nonparametric hypothesis tests are used to test whether differences in performance are statistically significant.

An Interpretive Constrained Linear Model for ResNet and MgNet ; We propose a constrained linear datafeaturemapping model as an interpretable mathematical model for image classification using a convolutional neural network CNN. From this viewpoint, we establish detailed connections between the traditional iterative schemes for linear systems and the architectures of the basic blocks of ResNet and MgNettype models. Using these connections, we present some modified ResNet models that compared with the original models have fewer parameters and yet can produce more accurate results, thereby demonstrating the validity of this constrained learning datafeaturemapping assumption. Based on this assumption, we further propose a general datafeature iterative scheme to show the rationality of MgNet. We also provide a systematic numerical study on MgNet to show its success and advantages in image classification problems and demonstrate its advantages in comparison with established networks.

VALSE A TaskIndependent Benchmark for Vision and Language Models Centered on Linguistic Phenomena ; We propose VALSE Vision And Language Structured Evaluation, a novel benchmark designed for testing generalpurpose pretrained vision and language VL models for their visiolinguistic grounding capabilities on specific linguistic phenomena. VALSE offers a suite of six tests covering various linguistic constructs. Solving these requires models to ground linguistic phenomena in the visual modality, allowing more finegrained evaluations than hitherto possible. We build VALSE using methods that support the construction of valid foils, and report results from evaluating five widelyused VL models. Our experiments suggest that current models have considerable difficulty addressing most phenomena. Hence, we expect VALSE to serve as an important benchmark to measure future progress of pretrained VL models from a linguistic perspective, complementing the canonical taskcentred VL evaluations.

Decomposing Natural Logic Inferences in Neural NLI ; In the interest of interpreting neural NLI models and their reasoning strategies, we carry out a systematic probing study which investigates whether these models capture the crucial semantic features central to natural logic monotonicity and concept inclusion. Correctly identifying valid inferences in downwardmonotone contexts is a known stumbling block for NLI performance, subsuming linguistic phenomena such as negation scope and generalized quantifiers. To understand this difficulty, we emphasize monotonicity as a property of a context and examine the extent to which models capture monotonicity information in the contextual embeddings which are intermediate to their decision making process. Drawing on the recent advancement of the probing paradigm, we compare the presence of monotonicity features across various models. We find that monotonicity information is notably weak in the representations of popular NLI models which achieve high scores on benchmarks, and observe that previous improvements to these models based on finetuning strategies have introduced stronger monotonicity features together with their improved performance on challenge sets.

Degenerate CahnHilliard and incompressible limit of a KellerSegel model ; The KellerSegel model is a wellknown system representing chemotaxis in living organisms. We study the convergence of a generalized nonlinear variant of the KellerSegel to the degenerate CahnHilliard system. This analysis is made possible from the observation that the KellerSegel system is equivalent to a relaxed version of the CahnHilliard system. Furthermore, this latter equivalent system has an interesting application in the modelling of living tissues. Indeed, compressible and incompressible porous medium type equations are widely used to describe the mechanical properties of living tissues. The relaxed degenerate CahnHilliard system, can be viewed as a compressible living tissue model for which the movement is driven by Darcy's law and takes into account the effects of the viscosity as well as surface tension at the surface of the tissue. We study the convergence of the KellerSegel system to the CahnHilliard equation and some of the analytical properties of the model such as the incompressible limit of our model. Our analysis relies on a priori estimates, compactness properties, and on the equivalence between the KellerSegel system and the relaxed degenerate CahnHilliard system.

Model order reduction strategies for weakly dispersive waves ; We focus on the numerical modelling of water waves by means of depth averaged models. We consider in particular PDE systems which consist in a nonlinear hyperbolic model plus a linear dispersive perturbation involving an elliptic operator. We propose two strategies to construct reduced order models for these problems, with the main focus being the control of the overhead related to the inversion of the elliptic operators, as well as the robustness with respect to variations of the flow parameters. In a first approach, only a linear reduction strategies is applied only to the elliptic component, while the computations of the nonlinear fluxes are still performed explicitly. This hybrid approach, referred to as pdROM, is compared to a hyperreduction strategy based on the empirical interpolation method to reduce also the nonlinear fluxes. We evaluate the two approaches on a variety of benchmarks involving a generalized variant of the BBMKdV model with a variable bottom, and a onedimensional enhanced weakly dispersive shallow water system. The results show the potential of both approaches in terms of cost reduction, with a clear advantage for the pdROM in terms of robustness, and for the EIMROM in terms of cost reduction.

Inflation with antisymmetric tensor field new candidates ; We study classes of inflation models driven by antisymmetric tensor field, with minimal and nonminimal couplings to gravity, that address known issues of such models considered in the past. First we show that with a different choice of the background structure of antisymmetric tensor field, inflation is supported even for the minimal model with quadratic potential contrary to past results. We also include the nonminimal coupling to gravity and analyse perturbations to the antisymmetric tensor as well as the tensor modes of perturbed metric. The two models differ in terms of the behaviour of tensor modes, where the speed of gravitational wave can be tuned to c in the latter model. The power spectrum and spectral index receive slight scale dependence. Finally, we consider a quartic potential motivated by the graceful exit to reheating phase, which requires a nonminimal coupling to support inflation. The two tensor modes of perturbed metric are found to evolve differently in this model, and give rise to a highly scaledependent power spectrum.

Fast and Accurate Prediction of Material Properties with ThreeBody TightBinding Model for the Periodic Table ; Parameterized tightbinding models fit to first principles calculations can provide an efficient and accurate quantum mechanical method for predicting properties of molecules and solids. However, welltested parameter sets are generally only available for a limited number of atom combinations, making routine use of this method difficult. Furthermore, most previous models consider only simple twobody interactions, which limits accuracy. To tackle these challenges, we develop a density functional theory database of nearly one million materials, which we use to fit a universal set of tightbinding parameters for 65 elements and their binary combinations. We include both twobody and threebody effective interaction terms in our model, plus selfconsistent charge transfer, enabling our model to work for metallic, covalent, and ionic bonds with the same parameter set. To ensure predictive power, we adopt a learning framework where we repeatedly test the model on new low energy crystal structures and then add them to the fitting dataset, iterating until predictions improve. We distribute the materials database and tools developed in this work publicly.

Gravitational Wave Imprints of LeftRight Symmetric Model with Minimal Higgs Sector ; We study the gravitational wave imprints of leftright symmetric model equipped with universal seesaw mechanism allowing for the natural generation of hierarchical masses of the Standard Model fermions. The scalar sector of this model is the minimal one, consisting of only two Higgs doublets. Following the construction of the full thermal potential for this model, we perform a scan of the entire parameter space and identify the region in which the cosmic phase transition associated with the leftright symmetry breaking gives gravitational wave signals detectable by a variety of planned spacebased interferometers. Then we also discuss the relevant collider implications of this beyond the Standard Model scenario.

Deep Filtering with DNN, CNN and RNN ; This paper is about a deep learning approach for linear and nonlinear filtering. The idea is to train a neural network with Monte Carlo samples generated from a nominal dynamic model. Then the network weights are applied to Monte Carlo samples from an actual dynamic model. A main focus of this paper is on the deep filters with three major neural network architectures DNN, CNN, RNN. Our deep filter compares favorably to the traditional Kalman filter in linear cases and outperform the extended Kalman filter in nonlinear cases. Then a switching model with jumps is studied to show the adaptiveness and power of our deep filtering. Among the three major NNs, the CNN outperform the others on average. while the RNN does not seem to be suitable for the filtering problem. One advantage of the deep filter is its robustness when the nominal model and actual model differ. The other advantage of deep filtering is real data can be used directly to train the deep neutral network. Therefore, model calibration can be bypassed all together.

Counterfactual Memorization in Neural Language Models ; Modern neural language models widely used in tasks across NLP risk memorizing sensitive information from their training data. As models continue to scale up in parameters, training data, and compute, understanding memorization in language models is both important from a learningtheoretical point of view, and is practically crucial in real world applications. An open question in previous studies of memorization in language models is how to filter out common memorization. In fact, most memorization criteria strongly correlate with the number of occurrences in the training set, capturing common memorization such as familiar phrases, public knowledge or templated texts. In this paper, we provide a principled perspective inspired by a taxonomy of human memory in Psychology. From this perspective, we formulate a notion of counterfactual memorization, which characterizes how a model's predictions change if a particular document is omitted during training. We identify and study counterfactuallymemorized training examples in standard text datasets. We further estimate the influence of each training example on the validation set and on generated texts, and show that this can provide direct evidence of the source of memorization at test time.

Accuracy and Application Scope Analysis for Linearized Branch Flow Model in Radial Distribution Systems ; An indepth analysis of linearized branch flow LBF model considering current injection and absolute value of impedance is proposed in this paper. The form of LBF model is based on two equations the current injection to meet KCL and the voltage drop to meet KVL. By representing the absolute value of complex load power with the current injection, LBF model is much simpler than alternating current power flow ACPF model. The results on theoretical analysis and numerical studies show that LBF exhibits the high accuracy in bus voltage magnitude but a poor performance in branch flow. Moreover, LBF is also compared with fast decoupled linearized power flow FDLPF model to verify its efficiency, thus proving its superiority for fast evaluation of largescale distribution systems with high accuracy in voltage magnitude. Finally, this paper analyzes three factors to lower LBF's errors of branch flow, as well as LBF's possible application scope.

Leaderless Consensus of Heterogeneous Multiple EulerLagrange Systems with Unknown Disturbance ; This paper studies the leaderless consensus problem of heterogeneous multiple networked EulerLagrange systems subject to persistent disturbances with unknown constant biases, amplitudes, initial phases, and frequencies. The main characteristic of this study is that none of the agents has information of a common reference model or of a common reference trajectory. Therefore, the agents must simultaneously and in a distributed way achieve consensus to a common reference model group model; achieve consensus to a common reference trajectory; and reject the unknown disturbances. We show that this is possible via a suitable combination of techniques of distributed observers', internal model principle, and adaptive regulation. The proposed design generalizes recent results on group model learning, which have been studied for linear agents over undirected networks. In this work, group model learning is achieved for EulerLagrange dynamics over directed networks in the presence of persistent unknown disturbances.

Extracting knowledge from features with multilevel abstraction ; Knowledge distillation aims at transferring the knowledge from a large teacher model to a small student model with great improvements of the performance of the student model. Therefore, the student network can replace the teacher network to deploy on lowresource devices since the higher performance, lower number of parameters and shorter inference time. Selfknowledge distillation SKD attracts a great attention recently that a student model itself is a teacher model distilling knowledge from. To the best of our knowledge, self knowledge distillation can be divided into two main streams data augmentation and refined knowledge auxiliary. In this paper, we purpose a novel SKD method in a different way from the main stream methods. Our method distills knowledge from multilevel abstraction features. Experiments and ablation studies show its great effectiveness and generalization on various kinds of tasks with various kinds of model structures. Our codes have been released on GitHub.

Relationship between the Euclidean and Lorentzian versions of the type IIB matrix model ; The type IIB matrix model was proposed as a nonperturbative formulation of superstring theory in 1996. We simulate a model that describes the late time behavior of the IIB matrix model by applying the complex Langevin method to overcome the sign problem. We clarify the relationship between the Euclidean and the Lorentzian versions of the type IIB matrix model in a recently discovered phase. By introducing a constraint, we obtain a model where the spacetime metric is Euclidean at early times, whereas it it dynamically becomes Lorentzian at late times.

Efficient Likelihoodbased Estimation via Annealing for Dynamic Structural Macrofinance Models ; Most solved dynamic structural macrofinance models are nonlinear andor nonGaussian statespace models with highdimensional and complex structures. We propose an annealed controlled sequential Monte Carlo method that delivers numerically stable and low variance estimators of the likelihood function. The method relies on an annealing procedure to gradually introduce information from observations and constructs globally optimal proposal distributions by solving associated optimal control problems that yield zero variance likelihood estimators. To perform parameter inference, we develop a new adaptive SMC2 algorithm that employs likelihood estimators from annealed controlled sequential Monte Carlo. We provide a theoretical stability analysis that elucidates the advantages of our methodology and asymptotic results concerning the consistency and convergence rates of our SMC2 estimators. We illustrate the strengths of our proposed methodology by estimating two popular macrofinance models a nonlinear new Keynesian dynamic stochastic general equilibrium model and a nonlinear nonGaussian consumptionbased longrun risk model.

Interpretable LowResource Legal Decision Making ; Over the past several years, legal applications of deep learning have been on the rise. However, as with other highstakes decision making areas, the requirement for interpretability is of crucial importance. Current models utilized by legal practitioners are more of the conventional machine learning type, wherein they are inherently interpretable, yet unable to harness the performance capabilities of datadriven deep learning models. In this work, we utilize deep learning models in the area of trademark law to shed light on the issue of likelihood of confusion between trademarks. Specifically, we introduce a modelagnostic interpretable intermediate layer, a technique which proves to be effective for legal documents. Furthermore, we utilize weakly supervised learning by means of a curriculum learning strategy, effectively demonstrating the improved performance of a deep learning model. This is in contrast to the conventional models which are only able to utilize the limited number of expensive manuallyannotated samples by legal experts. Although the methods presented in this work tackles the task of risk of confusion for trademarks, it is straightforward to extend them to other fields of law, or more generally, to other similar highstakes application scenarios.

FiniteWord Hyperlanguages ; Formal languages are in the core of models of computation and their behavior. A rich family of models for many classes of languages have been widely studied. Hyperproperties lift conventional tracebased languages from a set of execution traces to a set of sets of executions. Hyperproperties have been shown to be a powerful formalism for expressing and reasoning about informationflow security policies and important properties of cyberphysical systems. Although there is an extensive body of work on formallanguage representation of trace properties, we currently lack such a general characterization for hyperproperties. We introduce hyperlanguages over finite words and models for expressing them. Essentially, these models express multiple words by using assignments to quantified word variables. Relying on the standard models for regular languages, we propose hyperregular expressions and finiteword hyperautomata NFH, for modeling the class of regular hyperlanguages. We demonstrate the ability of regular hyperlanguages to express hyperproperties for finite traces. We explore the closure properties and the complexity of the fundamental decision problems such as nonemptiness, universality, membership, and containment for various fragments of NFH.

Fortunately, Discourse Markers Can Enhance Language Models for Sentiment Analysis ; In recent years, pretrained language models have revolutionized the NLP world, while achieving state of the art performance in various downstream tasks. However, in many cases, these models do not perform well when labeled data is scarce and the model is expected to perform in the zero or few shot setting. Recently, several works have shown that continual pretraining or performing a second phase of pretraining intertraining which is better aligned with the downstream task, can lead to improved results, especially in the scarce data setting. Here, we propose to leverage sentimentcarrying discourse markers to generate largescale weaklylabeled data, which in turn can be used to adapt language models for sentiment analysis. Extensive experimental results show the value of our approach on various benchmark datasets, including the finance domain. Code, models and data are available at httpsgithub.comibmtslmdiscoursemarkers.

Learning Sample Importance for CrossScenario Video Temporal Grounding ; The task of temporal grounding aims to locate video moment in an untrimmed video, with a given sentence query. This paper for the first time investigates some superficial biases that are specific to the temporal grounding task, and proposes a novel targeted solution. Most alarmingly, we observe that existing temporal ground models heavily rely on some biases e.g., high preference on frequent concepts or certain temporal intervals in the visual modal. This leads to inferior performance when generalizing the model in crossscenario test setting. To this end, we propose a novel method called Debiased Temporal Language Localizer DebiasTLL to prevent the model from naively memorizing the biases and enforce it to ground the query sentence based on true intermodal relationship. DebiasTLL simultaneously trains two models. By our design, a large discrepancy of these two models' predictions when judging a sample reveals higher probability of being a biased sample. Harnessing the informative discrepancy, we devise a data reweighing scheme for mitigating the data biases. We evaluate the proposed model in crossscenario temporal grounding, where the train test data are heterogeneously sourced. Experiments show largemargin superiority of the proposed method in comparison with stateoftheart competitors.

Enhanced total variation minimization for stable image reconstruction ; The total variation TV regularization has phenomenally boosted various variational models for image processing tasks. We propose to combine the backward diffusion process in the earlier literature of image enhancement with the TV regularization, and show that the resulting enhanced TV minimization model is particularly effective for reducing the loss of contrast. The main purpose of this paper is to establish stable reconstruction guarantees for the enhanced TV model from noisy subsampled measurements with two sampling strategies, nonadaptive sampling for general linear measurements and variabledensity sampling for Fourier measurements. In particular, under some weaker restricted isometry property conditions, the enhanced TV minimization model is shown to have tighter reconstruction error bounds than various TVbased models for the scenario where the level of noise is significant and the amount of measurements is limited. Advantages of the enhanced TV model are also numerically validated by preliminary experiments on the reconstruction of some synthetic, natural, and medical images.

A Likelihood Ratio based Domain Adaptation Method for E2E Models ; Endtoend E2E automatic speech recognition models like Recurrent Neural Networks Transducer RNNT are becoming a popular choice for streaming ASR applications like voice assistants. While E2E models are very effective at learning representation of the training data they are trained on, their accuracy on unseen domains remains a challenging problem. Additionally, these models require paired audio and text training data, are computationally expensive and are difficult to adapt towards the fast evolving nature of conversational speech. In this work, we explore a contextual biasing approach using likelihoodratio that leverages text data sources to adapt RNNT model to new domains and entities. We show that this method is effective in improving rare words recognition, and results in a relative improvement of 10 in 1best word error rate WER and 10 in nbest Oracle WER n8 on multiple outofdomain datasets without any degradation on a general dataset. We also show that complementing the contextual biasing adaptation with adaptation of a secondpass rescoring model gives additive WER improvements.

Neutrino seesaw models at oneloop matching Discrimination by effective operators ; Using the functional method, oneloop matching of the typeI, II and III seesaw models are investigated and the results are presented in both the Green's and the Warsaw bases. Although these models generate the same dimension5 Weinberg operator, they could induce quite different types of dimension6 effective operators that can be utilized for model discrimination. We also find the threshold effects from oneloop matching could be significant, which turn out to allow triggering electroweak symmetry breaking radiatively in typeII seesaw while forbid that in typeIIII models. An analytical criterion for such radiative symmetry breaking is also derived in typeII seesaw. Finally, we investigate the indirect signatures from different types of dimension6 operators at highenergy colliders, lowenergy precision experiments and forward physics facilities for model discrimination.

Channel Modeling in RISEmpowered Wireless Communications ; One of the most critical aspects of enabling nextgeneration wireless technologies is developing an accurate and consistent channel model to be validated effectively with the help of realworld measurements. From this point of view, remarkable research has recently been conducted to model propagation channels involving the modification of the wireless propagation environment through the inclusion of reconfigurable intelligent surfaces RISs. This study mainly aims to present a vision on channel modeling strategies for the RISempowered communications systems considering the stateoftheart channel and propagation modeling efforts in the literature. Moreover, it is also desired to draw attention to opensource and standardcompliant physical channel modeling efforts to provide comprehensive insights regarding the practical usecases of RISs in future wireless networks.

Exploit Customer Lifetime Value with Memoryless Experiments ; As a measure of the longterm contribution produced by customers in a service or product relationship, lifetime value, or LTV, can more comprehensively find the optimal strategy for service delivery. However, it is challenging to accurately abstract the LTV scene, model it reasonably, and find the optimal solution. The current theories either cannot precisely express LTV because of the single modeling structure, or there is no efficient solution. We propose a general LTV modeling method, which solves the problem that customers' longterm contribution is difficult to quantify while existing methods, such as modeling the clickthrough rate, only pursue the shortterm contribution. At the same time, we also propose a fast dynamic programming solution based on a mutated bisection method and the memoryless repeated experiments assumption. The model and method can be applied to different service scenarios, such as the recommendation system. Experiments on realworld datasets confirm the effectiveness of the proposed model and optimization method. In addition, this whole LTV structure was deployed at a large Ecommerce mobile phone application, where it managed to select optimal push message sending time and achieved a 10 LTV improvement.

Dilated Convolutional Neural Networks for Lightweight Diacritics Restoration ; Diacritics restoration has become a ubiquitous task in the Latinalphabetbased Englishdominated Internet language environment. In this paper, we describe a small footprint 1D dilated convolutionbased approach which operates on a characterlevel. We find that solutions based on 1D dilated convolutional neural networks are competitive alternatives to models based on recursive neural networks or linguistic modeling for the task of diacritics restoration. Our solution surpasses the performance of similarly sized models and is also competitive with larger models. A special feature of our solution is that it even runs locally in a web browser. We also provide a working example of this browserbased implementation. Our model is evaluated on different corpora, with emphasis on the Hungarian language. We performed comparative measurements about the generalization power of the model in relation to three Hungarian corpora. We also analyzed the errors to understand the limitation of corpusbased selfsupervised training.

A computational macroscale model for the time fractional poroelasticity problem in fractured and heterogeneous media ; In this work, we introduce a time memory formalism in poroelasticity model that couples the pressure and displacement. We assume this multiphysics process occurs in multicontinuum media. The mathematical model contains a coupled system of equations for pressures in each continuum and elasticity equations for displacements of the medium. We assume that the temporal dynamics is governed by fractional derivatives following some works in the literature. We derive an implicit finite difference approximation for time discretization based on the Caputo time fractional derivative. A Discrete Fracture Model DFM is used to model fluid flow through fractures and treat the complex network of fractures. We assume different fractional powers in fractures and matrix due to slow and fast dynamics. We develop a coarse grid approximation based on the Generalized Multiscale Finite Element Method GMsFEM, where we solve local spectral problems for construction of the multiscale basis functions. We present numerical results for the twodimensional model problems in fractured heterogeneous porous media. We investigate error analysis between reference finescale solution and multiscale solution with different numbers of multiscale basis functions. The results show that the proposed method can provide good accuracy on a coarse grid.

Reduced models for ETG transport in the pedestal ; This paper reports on the development of reduced models for electron temperature gradient ETG driven transport in the pedestal. Model development is enabled by a set of 61 nonlinear gyrokinetic simulations with input parameters taken from the pedestals in a broad range of experimental scenarios. The simulation data has been consolidated in a new database for gyrokinetic simulation data, the Multiscale Gyrokinetic Database MGKDB, facilitating the analysis. The modeling approach may be considered a generalization of the standard quasilinear mixing length procedure. The parameter eta, the ratio of the density to temperature gradient scale length, emerges as the key parameter for formulating an effective saturation rule. With a single orderunity fitting coefficient, the model achieves an RMS error of 15. A similar model for ETG particle flux is also described. We also present simple algebraic expressions for the transport informed by an algorithm for symbolic regression.

Multivariate sensitivity analysis for a largescale climate impact and adaptation model ; We develop a new efficient methodology for Bayesian global sensitivity analysis for largescale multivariate data. The focus is on computationally demanding models with correlated variables. A multivariate Gaussian process is used as a surrogate model to replace the expensive computer model. To improve the computational efficiency and performance of the model, compactly supported correlation functions are used. The goal is to generate sparse matrices, which give crucial advantages when dealing with large datasets, where we use crossvalidation to determine the optimal degree of sparsity. This method was combined with a robust adaptive Metropolis algorithm coupled with a parallel implementation to speed up the convergence to the target distribution. The method was applied to a multivariate dataset from the IMPRESSIONS Integrated Assessment Platform IAP2, an extension of the CLIMSAVE IAP, which has been widely applied in climate change impact, adaptation and vulnerability assessments. Our empirical results on synthetic and IAP2 data show that the proposed methods are efficient and accurate for global sensitivity analysis of complex models.

Exact Inference for Stochastic Epidemic Models via Uniformly Ergodic Block Sampling ; Stochastic epidemic models provide an interpretable probabilistic description of the spread of a disease through a population. Yet, fitting these models to partially observed data is a notoriously difficult task due to intractability of the likelihood for many classical models. To remedy this issue, this article introduces a novel dataaugmented MCMC algorithm for exact Bayesian inference under the stochastic SIR model, given only discretely observed counts of infection. In a MetropolisHastings step, the latent data are jointly proposed from a surrogate process carefully designed to closely resemble the SIR model, from which we can efficiently generate epidemics consistent with the observed data. This yields a method that explores the highdimensional latent space efficiently, and scales to outbreaks with hundreds of thousands of individuals. We show that the Markov chain underlying the algorithm is uniformly ergodic, and validate its performance via thorough simulation experiments and a case study on the 20132015 outbreak of Ebola Haemorrhagic Fever in Western Africa.

Exponential ergodicity for a stochastic twolayer quasigeostrophic model ; Ergodic properties of a stochastic medium complexity model for atmosphere and ocean dynamics are analysed. More specifically, a twolayer quasigeostrophic model for geophysical flows is studied, with the upper layer being perturbed by additive noise. This model is popular in the geosciences, for instance to study the effects of a stochastic wind forcing on the ocean. A rigorous mathematical analysis however meets with the challenge that in the model under study, the noise configuration is spatially degenerate as the stochastic forcing acts only on the top layer. Exponential convergence of solutions laws to the invariant measure is established, implying a spectral gap of the associated Markov semigroup on a space of Holder continuous functions. The approach provides a general framework for generalised coupling techniques suitable for applications to dissipative SPDEs. In case of the twolayer quasigeostrophic model, the results require the second layer to obey a certain passivity condition.

Classification Of Fake News Headline Based On Neural Networks ; Over the last few years, Text classification is one of the fundamental tasks in natural language processing NLP in which the objective is to categorize text documents into one of the predefined classes. The news is full of our life. Therefore, news headlines classification is a crucial task to connect users with the right news. The news headline classification is a kind of text classification, which can be generally divided into three mainly parts feature extraction, classifier selection, and evaluations. In this article, we use the dataset, containing news over a period of eighteen years provided by Kaggle platform to classify news headlines. We choose TFIDF to extract features and neural network as the classifier, while the evaluation metrics is accuracy. From the experiment result, it is obvious that our NN model has the best performance among these models in the metrics of accuracy. The higher the accuracy is, the better performance the model will gain. Our NN model owns the accuracy 0.8622, which is highest accuracy among these four models. And it is 0.0134, 0.033, 0.080 higher than its of other models.

A Framework for the HighLevel Specification and Verification of Synchronous Digital Logic Systems ; A syntactic model is presented for the specification of finitestate synchronous digital logic systems with complex inputoutput interfaces, which control the flow of data between opaque computational elements, and for the composition of compatible systems to form closedloop systems with no inputs or outputs. This model improves upon similar existing models with a novel approach to specifying input and output ports in a way which is uniform and symmetric. An automaton model is also presented for encoding arbitrary computational processes, and an algorithm is presented to generate an automaton representation of a closedloop system. Using the automaton model, the problem of timingagnostic verification of closedloop systems against a desired behavioural specification, encoded as the similarity of closedloop systems in terms of the set of computations performed, is shown to be decidable. The relationship between the models and realworld implementations of systems is discussed.

Certifying Model Accuracy under Distribution Shifts ; Certified robustness in machine learning has primarily focused on adversarial perturbations of the input with a fixed attack budget for each point in the data distribution. In this work, we present provable robustness guarantees on the accuracy of a model under bounded Wasserstein shifts of the data distribution. We show that a simple procedure that randomizes the input of the model within a transformation space is provably robust to distributional shifts under the transformation. Our framework allows the datumspecific perturbation size to vary across different points in the input distribution and is general enough to include fixedsized perturbations as well. Our certificates produce guaranteed lower bounds on the performance of the model for any natural or adversarial shift of the input distribution within a Wasserstein ball around the original distribution. We apply our technique to i certify robustness against natural nonadversarial transformations of images such as color shifts, hue shifts and changes in brightness and saturation, ii certify robustness against adversarial shifts of the input distribution, and iii show provable lower bounds hardness results on the performance of models trained on socalled unlearnable datasets that have been poisoned to interfere with model training.

Recycling Model Updates in Federated Learning Are Gradient Subspaces LowRank ; In this paper, we question the rationale behind propagating large numbers of parameters through a distributed system during federated learning. We start by examining the rank characteristics of the subspace spanned by gradients across epochs i.e., the gradientspace in centralized model training, and observe that this gradientspace often consists of a few leading principal components accounting for an overwhelming majority 9599 of the explained variance. Motivated by this, we propose the Lookback Gradient Multiplier LBGM algorithm, which exploits this lowrank property to enable gradient recycling between model update rounds of federated learning, reducing transmissions of large parameters to single scalars for aggregation. We analytically characterize the convergence behavior of LBGM, revealing the nature of the tradeoff between communication savings and model performance. Our subsequent experimental results demonstrate the improvement LBGM obtains in communication overhead compared to conventional federated learning on several datasets and deep learning models. Additionally, we show that LBGM is a general plugandplay algorithm that can be used standalone or stacked on top of existing sparsification techniques for distributed model training.

Parameters or Privacy A Provable Tradeoff Between Overparameterization and Membership Inference ; A surprising phenomenon in modern machine learning is the ability of a highly overparameterized model to generalize well small error on the test data even when it is trained to memorize the training data zero error on the training data. This has led to an arms race towards increasingly overparameterized models c.f., deep learning. In this paper, we study an underexplored hidden cost of overparameterization the fact that overparameterized models may be more vulnerable to privacy attacks, in particular the membership inference attack that predicts the potentially sensitive examples used to train a model. We significantly extend the relatively few empirical results on this problem by theoretically proving for an overparameterized linear regression model in the Gaussian data setting that membership inference vulnerability increases with the number of parameters. Moreover, a range of empirical studies indicates that more complex, nonlinear models exhibit the same behavior. Finally, we extend our analysis towards ridgeregularized linear regression and show in the Gaussian data setting that increased regularization also increases membership inference vulnerability in the overparameterized regime.

A discussion of stochastic dominance and meanrisk optimal portfolio problems based on meanvariancemixture models ; The classical Markowitz meanvariance model uses variance as a risk measure and calculates frontier portfolios in closed form by using standard optimization techniques. For general meanrisk models such closed form optimal portfolios are difficult to obtain. In this note, we obtain closed form expression for frontier portfolios under meanrisk criteria when risk is modelled by any finite lawinvariant convex measures of risk and when return vectors follow the class of normal meanvariance mixture NMVM distributions. To achieve this goal, we first present necessary as well as sufficient conditions for stochastic dominance within the class of one dimensional NMVM models and then we apply them to portfolio optimization problems. Our main result in this paper states that when return vectors follow the class of NMVM distributions the associated meanrisk frontier portfolios can be obtained by optimizing a Markowitz meanvariance model with an appropriately adjusted return vector.

Efficient Adapter Transfer of SelfSupervised Speech Models for Automatic Speech Recognition ; Selfsupervised learning SSL is a powerful tool that allows learning of underlying representations from unlabeled data. Transformer based models such as wav2vec 2.0 and HuBERT are leading the field in the speech domain. Generally these models are finetuned on a small amount of labeled data for a downstream task such as Automatic Speech Recognition ASR. This involves retraining the majority of the model for each task. Adapters are small lightweight modules which are commonly used in Natural Language Processing NLP to adapt pretrained models to new tasks. In this paper we propose applying adapters to wav2vec 2.0 to reduce the number of parameters required for downstream ASR tasks, and increase scalability of the model to multiple tasks or languages. Using adapters we can perform ASR while training fewer than 10 of parameters per task compared to full finetuning with little degradation of performance. Ablations show that applying adapters into just the top few layers of the pretrained network gives similar performance to full transfer, supporting the theory that higher pretrained layers encode more phonemic information, and further optimizing efficiency.

Do Language Models Learn PositionRole Mappings ; How is knowledge of positionrole mappings in natural language learned We explore this question in a computational setting, testing whether a variety of wellperforming pertained language models BERT, RoBERTa, and DistilBERT exhibit knowledge of these mappings, and whether this knowledge persists across alternations in syntactic, structural, and lexical alternations. In Experiment 1, we show that these neural models do indeed recognize distinctions between theme and recipient roles in ditransitive constructions, and that these distinct patterns are shared across construction type. We strengthen this finding in Experiment 2 by showing that finetuning these language models on novel theme and recipientlike tokens in one paradigm allows the models to make correct predictions about their placement in other paradigms, suggesting that the knowledge of these mappings is shared rather than independently learned. We do, however, observe some limitations of this generalization when tasks involve constructions with novel ditransitive verbs, hinting at a degree of lexical specificity which underlies model performance.

Robust Hybrid Learning With Expert Augmentation ; Hybrid modelling reduces the misspecification of expert models by combining them with machine learning ML components learned from data. Similarly to many ML algorithms, hybrid model performance guarantees are limited to the training distribution. Leveraging the insight that the expert model is usually valid even outside the training domain, we overcome this limitation by introducing a hybrid data augmentation strategy termed textitexpert augmentation. Based on a probabilistic formalization of hybrid modelling, we demonstrate that expert augmentation, which can be incorporated into existing hybrid systems, improves generalization. We empirically validate the expert augmentation on three controlled experiments modelling dynamical systems with ordinary and partial differential equations. Finally, we assess the potential realworld applicability of expert augmentation on a dataset of a real double pendulum.

The Lifecycle of a Statistical Model Model Failure Detection, Identification, and Refitting ; The statistical machine learning community has demonstrated considerable resourcefulness over the years in developing highly expressive tools for estimation, prediction, and inference. The bedrock assumptions underlying these developments are that the data comes from a fixed population and displays little heterogeneity. But reality is significantly more complex statistical models now routinely fail when released into realworld systems and scientific applications, where such assumptions rarely hold. Consequently, we pursue a different path in this paper visavis the wellworn trail of developing new methodology for estimation and prediction. In this paper, we develop tools and theory for detecting and identifying regions of the covariate space subpopulations where model performance has begun to degrade, and study intervening to fix these failures through refitting. We present empirical results with three realworld data sets including a time series involving forecasting the incidence of COVID19 showing that our methodology generates interpretable results, is useful for tracking model performance, and can boost model performance through refitting. We complement these empirical results with theory proving that our methodology is minimax optimal for recovering anomalous subpopulations as well as refitting to improve accuracy in a structured normal means setting.

Modeling bacterial flagellar motor with new structure information Rotational dynamics of two interacting protein nanorings ; In this article, we develop a mathematical model for the rotary bacterial flagellar motor BFM based on the recently discovered structure of the stator complex MotA5MotB2. The structure suggested that the stator also rotates. The BFM is modeled as two rotating nanorings that interact with each other. Specifically, translocation of protons through the stator complex drives rotation of the MotA pentamer ring, which in turn drives rotation of the FliG ring in the rotor via interactions between the MotA ring of the stator and the FliG ring of the rotor. Preliminary results from the structureinformed model are consistent with the observed torquespeed relation. More importantly, the model predicts distinctive rotor and stator dynamics and their load dependence, which may be tested by future experiments. Possible approaches to verify and improve the model to further understanding of the molecular mechanism for torque generation in BFM are also discussed.

Pleasant behavior of swampland conjectures in the face of specific inflationary models ; Recently researchers have been studying various conditions as swampland criteria in cosmological implications. They have studied many inflation models with different swampland conditions. Occasionally these conjectures are modified and lead to fascinating points. The swampland conjecture is in contradiction with the slowroll singlefield inflation model. So in this paper, we will briefly describe resolving this contradiction with the special method; we consider a small parameter lambda with respect to a coupling function in inflation and resolve this important antithesis. Then we will discuss a new point of view in studying inflation models according to the swampland criteria. Therefore we introduce some inflation models and challenge the swampland criteria. We specify the allowable range of cosmological parameters as the tensortoscalar ratio r the scalar spectral index ns according to the latest observable data. Next, we study the new constraints to determine the compatibility or incompatibility of the corresponding model with the swampland conjectures. Finally, we express the results and compare them with the latest observational data by plotting some figures.

Coefficient Decomposition of Spatial Regressive Models Based on Standardized Variables ; Spatial autocorrelation analysis is the basis for spatial autoregressive modeling. However, the relationships between spatial correlation coefficients and spatial regression models are not yet well clarified. The paper is devoted to explore the deep structure of spatial regression coefficients. By means of mathematical reasoning, a pair of formulae of canonical spatial regression coefficients are derived from a general spatial regression model based on standardized variables. The spatial auto and lagregression coefficients are reduced to a series of statistic parameters and measurements, including conventional regressive coefficient, Pearson correlation coefficient, Moran's indexes, spatial crosscorrelation coefficients, and the variance of prediction residuals. The formulae show determinate inherent relationships between spatial correlation coefficients and spatial regression coefficients. New finding is as below the spatial autoregressive coefficient mainly depends on the Moran's index of the independent variable, while the spatial lagregressive coefficient chiefly depends on the crosscorrelation coefficient of independent variable and dependent variable. The observational data of an urban system in Beijing, Tianjin, and Hebei region of China were employed to verify the newly derived formulae, and the results are satisfying. The new formulae and their variates are helpful for understand spatial regression models from the perspective of spatial correlation and can be used to assist spatial regression modeling.

Interacting and Noninteracting Renyi Holographic Dark Energy Models in DGP Braneworld ; We investigate both the interacting and noninteracting R'enyi Holographic Dark Energy RHDE models in DGP brane world framework. Cosmological parameters and their evolutions are probed to obtain realistic cosmological models. We note that both the models accommodate the present accelerating phase of expansion with the observed dark energy density. Classical stability of the cosmological model and Om diagnostic are also studied to test the suitability of the cosmological models obtained in the presence of RHDE in DGP braneworld.

Simple Models and Biased Forecasts ; This paper proposes a framework in which agents are constrained to use simple timeseries models to forecast economic variables and characterizes the resulting biases. It considers agents who can only entertain statespace models with no more than d states, where d measures the agents' cognitive abilities. When the true datagenerating process does not have a dstate representation, agents end up with misspecified models and biased forecasts. Under some assumptions, agents attend to the most persistent observables at the expense of less persistent ones. This bias anchors agents' forwardlooking decisions to persistent state variables and increases comovement among those decisions. The paper proceeds to study the implications of the theory in the context of newKeynesian, real business cycle, and DiamondMortensenPissarides models. In each case, constraining agents to use simple models brings the outcomes more in line with stylized facts.