Datasets:

sswt

/

arxiv_sample_50K

like 0

Asymptotic behaviors for Blackstock's model of thermoviscous flow ; We study a fundamental model in nonlinear acoustics, precisely, the general Blackstock's model that is, without Becker's assumption in the whole space mathbbRn. This model describes nonlinear acoustics in perfect gases under the irrotational flow. By means of the Fourier analysis we will derive L2 estimates for the solution of the linear homogeneous problem and its derivatives. Then, we will apply these estimates to study three different topics the optimality of the decay estimates in the case ngeqslant 5 and the optimal growth rate for the L2norm of the solution for n3,4; the singular limit problem in determining the first and secondorder profiles for the solution of the linear Blackstock's model with respect to the small thermal diffusivity; the proof of the existence of global in time small data Sobolev solutions with suitable regularity for a nonlinear Blackstock's model.

Yukawa coupling unification in an mathsfSO10 model consistent with Fermilab g2 result ; We investigate the Yukawa coupling unification for the third generation in a class of mathsfSO10 unified models which are consistent with the 4.2 sigma deviation from the standard model of the muon g2 seen by the Fermilab experiment E989. A recent analysis in supergravity grand unified models shows that such an effect can arise from supersymmetric loops correction. Using a neural network, we further analyze regions of the parameter space where Yukawa coupling unification consistent with the Fermilab result can appear. In the analysis we take into account the contributions to Yukawas from the cubic and the quartic interactions. We test the model at the high luminosity and high energy LHC and estimate the integrated luminosities needed to discover sparticles predicted by the model.

Learning to Learn to be Right for the Right Reasons ; Improving model generalization on heldout data is one of the core objectives in commonsense reasoning. Recent work has shown that models trained on the dataset with superficial cues tend to perform well on the easy test set with superficial cues but perform poorly on the hard test set without superficial cues. Previous approaches have resorted to manual methods of encouraging models not to overfit to superficial cues. While some of the methods have improved performance on hard instances, they also lead to degraded performance on easy instances. Here, we propose to explicitly learn a model that does well on both the easy test set with superficial cues and hard test set without superficial cues. Using a metalearning objective, we learn such a model that improves performance on both the easy test set and the hard test set. By evaluating our models on Choice of Plausible Alternatives COPA and Commonsense Explanation, we show that our proposed method leads to improved performance on both the easy test set and the hard test set upon which we observe up to 16.5 percentage points improvement over the baseline.

Radiationreaction force and multipolar waveforms for eccentric, spinaligned binaries in the effectiveonebody formalism ; While most binary inspirals are expected to have circularized before they enter the LIGOVirgo frequency band, a small fraction of those binaries could have nonnegligible orbital eccentricity depending on their formation channel. Hence, it is important to accurately model eccentricity effects in waveform models used to detect those binaries, infer their properties, and shed light on their astrophysical environment. We develop a multipolar effectiveonebody EOB eccentric waveform model for compact binaries whose components have spins aligned or antialigned with the orbital angular momentum. The waveform model contains eccentricity effects in the radiationreaction force and gravitational modes through second postNewtonian PN order, including tail effects, and spinorbit and spinspin couplings. We recast the PNexpanded, eccentric radiationreaction force and modes in factorized form so that the newly derived terms can be directly included in the stateoftheart, quasicircularorbit EOB model currently used in LIGOVirgo analyses i.e., the SEOBNRv4HM model.

Safe Chance Constrained Reinforcement Learning for Batch Process Control ; Reinforcement Learning RL controllers have generated excitement within the control community. The primary advantage of RL controllers relative to existing methods is their ability to optimize uncertain systems independently of explicit assumption of process uncertainty. Recent focus on engineering applications has been directed towards the development of safe RL controllers. Previous works have proposed approaches to account for constraint satisfaction through constraint tightening from the domain of stochastic model predictive control. Here, we extend these approaches to account for plantmodel mismatch. Specifically, we propose a datadriven approach that utilizes Gaussian processes for the offline simulation model and use the associated posterior uncertainty prediction to account for joint chance constraints and plantmodel mismatch. The method is benchmarked against nonlinear model predictive control via case studies. The results demonstrate the ability of the methodology to account for process uncertainty, enabling satisfaction of joint chance constraints even in the presence of plantmodel mismatch.

Regression Modeling for Recurrent Events Using R Package reReg ; Recurrent event analyses have found a wide range of applications in biomedicine, public health, and engineering, among others, where study subjects may experience a sequence of event of interest during followup. The R package reReg Chiou and Huang 2021 offers a comprehensive collection of practical and easytouse tools for regression analysis of recurrent events, possibly with the presence of an informative terminal event. The regression framework is a general scalechange model which encompasses the popular Coxtype model, the accelerated rate model, and the accelerated mean model as special cases. Informative censoring is accommodated through a subjectspecific frailty without no need for parametric specification. Different regression models are allowed for the recurrent event process and the terminal event. Also included are visualization and simulation tools.

Cosmological dynamics of interacting dark energy and dark matter in viable models of fR gravity ; In this work, we investigate the dynamics of the interacting dark energy and dark matter in viable models of fR gravity by using a standard framework of dynamical system analysis. A simple form of the interacting dark energy Q3alpha Hrhom is used to study three viable models of fR gravity which are consistent with local gravity constraints and satisfying conditions for the cosmological viability. As a result, we find that the fixed points are slightly modified from those obtained in the standard noninteracting analysis of fR gravity proposed so far in the literature. In our models of adding this interaction, we find that the dynamical profiles of the universe in the viable fR dark energy models are modified by the interaction term as well as their relevant model parameters. Moreover, our results yield the correct cosmological evolution with additional constraint parameter, alpha, from the interacting dark energy.

On dependency models and dependent generalized sensitivity indices ; In this paper, we derive copulabased and empirical dependency models DMs for simulating nonindependent variables, and then propose a new way for determining the distribution of the model outputs conditional on every subset of inputs. Our approach relies on equivalent representations of the model outputs using DMs, and an algorithm is provided for selecting the representations that are necessary and sufficient for deriving the distribution of the model outputs given each subset of inputs. In sensitivity analysis, the selected representations allow for assessing the main, total and interactions effects of each subset of inputs. We then introduce the firstorder and total indices of every subset of inputs with the former index less than the latter. Consistent estimators of such indices are provided, including their asymptotic distributions. Analytical and numerical results are provided using single and multivariate response models.

Analyses of scalar potential and lepton flavor violating decays in a model with A4 symmetry ; We have considered a model, originally proposed by Ma and Wegman, where the mixing pattern in neutrino sector is explained with three Higgs doublets, six Higgs triplets and A4 symmetry. The mixing pattern is explained with the help of vacuum expectation values VEVs of the above mentioned doublets and triplets. In order to study about the VEVs of the scalar fields, we construct the full invariant scalar potential of this model. After minimizing this scalar potential, we have found that two Higgs triplets can acquire zero VEVs. In order to generate nonzero VEVs to all the six Higgs triplets, we have added two more Higgs doublets to the model. Thereafter we have demonstrated that the current neutrino oscillation data can be consistently explained in our model. To study some phenomenological implications of this model, we have worked out on the branching ratios for lepton flavor violating decays.

Efficient Pretrained Features and Recurrent PseudoLabeling in Unsupervised Domain Adaptation ; Domain adaptation DA mitigates the domain shift problem when transferring knowledge from one annotated domain to another similar but different unlabeled domain. However, existing models often utilize one of the ImageNet models as the backbone without exploring others, and finetuning or retraining the backbone ImageNet model is also timeconsuming. Moreover, pseudolabeling has been used to improve the performance in the target domain, while how to generate confident pseudo labels and explicitly align domain distributions has not been well addressed. In this paper, we show how to efficiently opt for the best pretrained features from seventeen wellknown ImageNet models in unsupervised DA problems. In addition, we propose a recurrent pseudolabeling model using the best pretrained features termed PRPL to improve classification performance. To show the effectiveness of PRPL, we evaluate it on three benchmark datasets, OfficeCaltech10, Office31, and OfficeHome. Extensive experiments show that our model reduces computation time and boosts the mean accuracy to 98.1, 92.4, and 81.2, respectively, substantially outperforming the state of the art.

Modeling Managerial Search Behavior based on Simon's Concept of Satisficing ; Computational models of managerial search often build on backwardlooking search based on hillclimbing algorithms. Regardless of its prevalence, there is some evidence that this family of algorithms does not universally represent managers' search behavior. Against this background, the paper proposes an alternative algorithm that captures key elements of Simon's concept of satisficing which received considerable support in behavioral experiments. The paper contrasts the satisficingbased algorithm to two variants of hillclimbing search in an agentbased model of a simple decisionmaking organization. The model builds on the framework of NK fitness landscapes which allows controlling for the complexity of the decision problem to be solved. The results suggest that the model's behavior may remarkably differ depending on whether satisficing or hillclimbing serves as an algorithmic representation for decisionmakers' search. Moreover, with the satisficing algorithm, results indicate oscillating aspiration levels, even to the negative, and intense and potentially destabilizing search activities when intraorganizational complexity increases. Findings may shed some new light on prior computational models of decisionmaking in organizations and point to avenues for future research.

Analytical formulation for multidimensional continuous opinion models ; Usually, opinion formation models assume that individuals have an opinion about a given topic which can change due to interactions with others. However, individuals can have different opinions in different topics and therefore ndimensional models are best suited to deal with these cases. While there have been many efforts to develop analytical models for one dimensional opinion models, less attention has been paid to multidimensional ones. In this work, we develop an analytical approach for multidimensional models of continuous opinions where dimensions can be correlated or uncorrelated. We show that for any generic reciprocal interactions between agents, the mean value of initial opinion distribution is conserved. Moreover, for positive social influence interaction mechanisms, the variance of opinion distributions decreases with time and the system converges to a delta distributed function. In particular, we calculate the convergence time when agents get closer in a discrete quantity after interacting, showing a clear difference between correlated and uncorrelated cases.

Deep Image Destruction Vulnerability of Deep ImagetoImage Models against Adversarial Attacks ; Recently, the vulnerability of deep image classification models to adversarial attacks has been investigated. However, such an issue has not been thoroughly studied for imagetoimage tasks that take an input image and generate an output image e.g., colorization, denoising, deblurring, etc. This paper presents comprehensive investigations into the vulnerability of deep imagetoimage models to adversarial attacks. For five popular imagetoimage tasks, 16 deep models are analyzed from various standpoints such as output quality degradation due to attacks, transferability of adversarial examples across different tasks, and characteristics of perturbations. We show that unlike image classification tasks, the performance degradation on imagetoimage tasks largely differs depending on various factors, e.g., attack methods and task objectives. In addition, we analyze the effectiveness of conventional defense methods used for classification models in improving the robustness of the imagetoimage models.

Local Asymptotic Mixed Normality via Transition Density Approximation and an Application to Ergodic JumpDiffusion Processes ; We study sufficient conditions for local asymptotic mixed normality. We weaken the sufficient conditions in Theorem 1 of Jeganathan Sankhya Ser. A 1982 so that they can be applied to a wider class of statistical models including a jumpdiffusion model. Moreover, we show that local asymptotic mixed normality of a statistical model generated by approximated transition density functions is implied for the original model. Together with density approximation by means of thresholding techniques, we show local asymptotic normality for a statistical model of discretely observed jumpdiffusion processes where the drift coefficient, diffusion coefficient, and jump structure are parametrized. As a consequence, the quasimaximumlikelihood and Bayestype estimators proposed in Shimizu and Yoshida Stat. Inference Stoch. Process. 2006 and Ogihara and Yoshida Stat. Inference Stoch. Process. 2011 are shown to be asymptotically efficient in this model. Moreover, we can construct asymptotically uniformly most powerful tests for the parameters.

MathBERT A PreTrained Model for Mathematical Formula Understanding ; Largescale pretrained models like BERT, have obtained a great success in various Natural Language Processing NLP tasks, while it is still a challenge to adapt them to the mathrelated tasks. Current pretrained models neglect the structural features and the semantic correspondence between formula and its context. To address these issues, we propose a novel pretrained model, namely textbfMathBERT, which is jointly trained with mathematical formulas and their corresponding contexts. In addition, in order to further capture the semanticlevel structural features of formulas, a new pretraining task is designed to predict the masked formula substructures extracted from the Operator Tree OPT, which is the semantic structural representation of formulas. We conduct various experiments on three downstream tasks to evaluate the performance of MathBERT, including mathematical information retrieval, formula topic classification and formula headline generation. Experimental results demonstrate that MathBERT significantly outperforms existing methods on all those three tasks. Moreover, we qualitatively show that this pretrained model effectively captures the semanticlevel structural information of formulas. To the best of our knowledge, MathBERT is the first pretrained model for mathematical formula understanding.

pyBKT An Accessible Python Library of Bayesian Knowledge Tracing Models ; Bayesian Knowledge Tracing, a model used for cognitive mastery estimation, has been a hallmark of adaptive learning research and an integral component of deployed intelligent tutoring systems ITS. In this paper, we provide a brief history of knowledge tracing model research and introduce pyBKT, an accessible and computationally efficient library of model extensions from the literature. The library provides data generation, fitting, prediction, and crossvalidation routines, as well as a simple to use data helper interface to ingest typical tutor log dataset formats. We evaluate the runtime with various dataset sizes and compare to past implementations. Additionally, we conduct sanity checks of the model using experiments with simulated data to evaluate the accuracy of its EM parameter learning and use realworld data to validate its predictions, comparing pyBKT's supported model variants with results from the papers in which they were originally introduced. The library is open source and open license for the purpose of making knowledge tracing more accessible to communities of research and practice and to facilitate progress in the field through easier replication of past approaches.

Absence of Phase Transition in Random Language Model ; The Random Language Model, proposed as a simple model of human languages, is defined by the averaged model of a probabilistic contextfree grammar. This grammar expresses the process of sentence generation as a tree graph with nodes having symbols as variables. Previous studies proposed that a phase transition, which can be considered to represent the emergence of order in language, occurs in the random language model. We discuss theoretically that the analysis of the order parameter introduced in previous studies can be reduced to solving the maximum eigenvector of the transition probability matrix determined by a grammar. This helps analyze the distribution of a quantity determining the behavior of the order parameter and reveals that no phase transition occurs. Our results suggest the need to study a more complex model such as a probabilistic contextsensitive grammar, in order for phase transitions to occur.

Goldilocks JustRight Tuning of BERT for TechnologyAssisted Review ; Technologyassisted review TAR refers to iterative active learning workflows for document review in high recall retrieval HRR tasks. TAR research and most commercial TAR software have applied linear models such as logistic regression to lexical features. Transformerbased models with supervised tuning are known to improve effectiveness on many text classification tasks, suggesting their use in TAR. We indeed find that the pretrained BERT model reduces review cost by 10 to 15 in TAR workflows simulated on the RCV1v2 newswire collection. In contrast, we likewise determined that linear models outperform BERT for simulated legal discovery topics on the Jeb Bush email collection. This suggests the match between transformer pretraining corpora and the task domain is of greater significance than generally appreciated. Additionally, we show that justright language model finetuning on the task collection before starting active learning is critical. Too little or too much finetuning hinders performance, worse than that of linear models, even for a favorable corpus such as RCV1v2.

Exploring Quantum Perceptron and Quantum Neural Network structures with a teacherstudent scheme ; Nearterm quantum devices can be used to build quantum machine learning models, such as quantum kernel methods and quantum neural networks QNN to perform classification tasks. There have been many proposals how to use variational quantum circuits as quantum perceptrons or as QNNs. The aim of this work is to systematically compare different QNN architectures and to evaluate their relative expressive power with a teacherstudent scheme. Specifically, the teacher model generates the datasets mapping random inputs to outputs which then have to be learned by the student models. This way, we avoid training on arbitrary data sets and allow to compare the learning capacity of different models directly via the loss, the prediction map, the accuracy and the relative entropy between the prediction maps. We focus particularly on a quantum perceptron model inspired by the recent work of Tacchino et. al. citeTacchino1 and compare it to the data reuploading scheme that was originally introduced by P'erezSalinas et. al. citedatareuploading. We discuss alterations of the perceptron model and the formation of deep QNN to better understand the role of hidden units and nonlinearities in these architectures.

On SIR epidemic models with feedbackcontrolled interactions and network effects ; We study extensions of the classical SIR model of epidemic spread. First, we consider a single population modified SIR epidemics model in which the contact rate is allowed to be an arbitrary function of the fraction of susceptible and infected individuals. This allows one to model either the reaction of individuals to the information about the spread of the disease or the result of government restriction measures, imposed to limit social interactions and contain contagion. We study the effect of both smooth dependancies of the contact rate for which we prove the existence of a threshold phenomenon that generalizes the wellknown dichotomy associated to the reproduction rate parameter in the classical SIR model, and discontinuous feedback terms, which can be studied using tools from sliding mode control. Finally, we consider network SIR models involving different subpopulations that interact on a contact graph and present some preliminary simulations of modified versions of the classic SIR network.

Adapting Monolingual Models Data can be Scarce when Language Similarity is High ; For many minority languages, the resources needed to train large models are not available. We investigate the performance of zeroshot transfer learning with as little data as possible, and the influence of language similarity in this process. We retrain the lexical layers of four BERTbased models using data from two lowresource target language varieties, while the Transformer layers are independently finetuned on a POStagging task in the model's source language. By combining the new lexical layers and finetuned Transformer layers, we achieve high task performance for both target languages. With high language similarity, 10MB of data appears sufficient to achieve substantial monolingual transfer performance. Monolingual BERTbased models generally achieve higher downstream task performance after retraining the lexical layer than multilingual BERT, even when the target language is included in the multilingual model.

A Metamodel Structure For Regression Analysis Application To Prediction Of Autism Spectrum Disorder Severity ; Traditional regression models do not generalize well when learning from small and noisy datasets. Here we propose a novel metamodel structure to improve the regression result. The metamodel is composed of multiple classification base models and a regression model built upon the base models. We test this structure on the prediction of autism spectrum disorder ASD severity as measured by the ADOS communication ADOS COMM score from restingstate fMRI data, using a variety of base models. The metamodel outperforms traditional regression models as measured by the Pearson correlation coefficient between true and predicted scores and stability. In addition, we found that the metamodel is more flexible and more generalizable.

SimJEB Simulated Jet Engine Bracket Dataset ; This paper introduces the Simulated Jet Engine Bracket Dataset SimJEB a new, public collection of crowdsourced mechanical brackets and accompanying structural simulations. SimJEB is applicable to a wide range of geometry processing tasks; the complexity of the shapes in SimJEB offer a challenge to automated geometry cleaning and meshing, while categorical labels and structural simulations facilitate classification and regression i.e. engineering surrogate modeling. In contrast to existing shape collections, SimJEB's models are all designed for the same engineering function and thus have consistent structural loads and support conditions. On the other hand, SimJEB models are more complex, diverse, and realistic than the synthetically generated datasets commonly used in parametric surrogate model evaluation. The designs in SimJEB were derived from submissions to the GrabCAD Jet Engine Bracket Challenge an open engineering design competition with over 700 handdesigned CAD entries from 320 designers representing 56 countries. Each model has been cleaned, categorized, meshed, and simulated with finite element analysis according to the original competition specifications. The result is a collection of 381 diverse, highquality and applicationfocused designs for advancing geometric deep learning, engineering surrogate modeling, automated cleaning and related geometry processing tasks.

An Intelligent Model for Solving Manpower Scheduling Problems ; The manpower scheduling problem is a critical research field in the resource management area. Based on the existing studies on scheduling problem solutions, this paper transforms the manpower scheduling problem into a combinational optimization problem under multiconstraint conditions from a new perspective. It also uses logical paradigms to build a mathematical model for problem solution and an improved multidimensional evolution algorithm for solving the model. Moreover, the constraints discussed in this paper basically cover all the requirements of human resource coordination in modern society and are supported by our experiment results. In the discussion part, we compare our model with other heuristic algorithms or linear programming methods and prove that the model proposed in this paper makes a 25.7 increase in efficiency and a 17 increase in accuracy at most. In addition, to the numerical solution of the manpower scheduling problem, this paper also studies the algorithm for scheduling task list generation and the method of displaying scheduling results. As a result, we not only provide various modifications for the basic algorithm to solve different condition problems but also propose a new algorithm that increases at least 28.91 in time efficiency by comparing with different baseline models.

An Efficient Bayes Coding Algorithm for the NonStationary Source in Which Context Tree Model Varies from Interval to Interval ; The context tree source is a source model in which the occurrence probability of symbols is determined from a finite past sequence, and is a broader class of sources that includes i.i.d. and Markov sources. The proposed source model in this paper represents that a subsequence in each interval is generated from a different context tree model. The Bayes code for such sources requires weighting of the posterior probability distributions for the change patterns of the context tree source and for all possible context tree models. Therefore, the challenge is how to reduce this exponential order computational complexity. In this paper, we assume a special class of prior probability distribution of change patterns and context tree models, and propose an efficient Bayes coding algorithm whose computational complexity is the polynomial order.

Cosmological Evolution via InteractingNonInteracting Holographic Dark Energy Model for Curved FLRW Spacetime in Rastall Gravity ; In this work, we explore the phenomenon of cosmic evolution using curved FLRW spacetime bounded by apparent horizon with a specific holographic cutoff. To this end, we use the framework of Rastall gravity RG and universe is assumed to be consists of interacting noninteracting dark energy DE and dark matter DM. In both scenarios, we evaluate exact solutions of the dynamical equations and constraint the holographic parameter c2z assuming a slowly varying function of redshift during noninteracting model. For interacting model, we consider c2 as a constant function of redshift. Moreover, we analyze nature of the obtained results via deceleration parameter q, statefinder pair j,s and Omzdiagnostic by constraining the involved model parameters using latest observational data. The graphical analysis showed that interacting model is very close to Lambda cold DM LambdaCDM model as compared to noninteracting case. We conclude that this holographic proposal is enough to describe the cosmic evolution at an accelerating rate.

SCAU Modeling spectral causality for multivariate time series with applications to electroencephalograms ; Electroencephalograms EEG are noninvasive measurement signals of electrical neuronal activity in the brain. One of the current major statistical challenges is formally measuring functional dependency between those complex signals. This paper, proposes the spectral causality model SCAU, a robust linear model, under a causality paradigm, to reflect inter and intrafrequency modulation effects that cannot be identifiable using other methods. SCAU inference is conducted with three main steps a signal decomposition into frequency bins, b intermediate spectral band mapping, and c dependency modeling through frequencyspecific autoregressive models VAR. We apply SCAU to study complex dependencies during visual and lexical fluency tasks word generation and visual fixation in 26 participants' EEGs. We compared the connectivity networks estimated using SCAU with respect to a VAR model. SCAU networks show a clear contrast for both stimuli while the magnitude links also denoted a low variance in comparison with the VAR networks. Furthermore, SCAU dependency connections not only were consistent with findings in the neuroscience literature, but it also provided further evidence on the directionality of the spatiospectral dependencies such as the deltaoriginated and thetainduced links in the frontotemporal brain network.

Bayesian inference under model misspecification using transportLagrangian distances an application to seismic inversion ; Model misspecification constitutes a major obstacle to reliable inference in many inverse problems. Inverse problems in seismology, for example, are particularly affected by misspecification of wave propagation velocities. In this paper, we focus on a specific seismic inverse problem fullwaveform moment tensor inversion and develop a Bayesian framework that seeks robustness to velocity misspecification. A novel element of our framework is the use of transportLagrangian TL distances between observed and model predicted waveforms to specify a loss function, and the use of this loss to define a generalized belief update via a Gibbs posterior. The TL distance naturally disregards certain features of the data that are more sensitive to model misspecification, and therefore produces less biased or dispersed posterior distributions in this setting. To make the latter notion precise, we use several diagnostics to assess the quality of inference and uncertainty quantification, i.e., continuous rank probability scores and rank histograms. We interpret these diagnostics in the Bayesian setting and compare the results to those obtained using more typical Gaussian noise models and squarederror loss, under various scenarios of misspecification. Finally, we discuss potential generalizability of the proposed framework to a broader class of inverse problems affected by model misspecification.

Distinguishing Dark Energy Models with Neutrino Oscillations ; Dark Energy models are numerous and distinguishing between them is becoming difficult. However, using distinct observational probes can ease this quest and gives better assessment to the nature of Dark energy. To this end, the plausibility of neutrino oscillations to be a probe of Dark Energy models is investigated. First, a generalized formalism of neutrino spinor field interaction with a classical scalar field in curved spacetime is presented. This formalism is then applied to two classes of Dark Energy models in a flat FriedmanLemaitreRobertsonWalker metric a Cosmological Constant and scalar field Dark Energy coupled to neutrinos. By looking at the neutrino oscillation probability's evolution with redshift, these models can be distinguished, for certain neutrino and scalar field coupling properties. This evolution could be traced by neutrino flux in future underground, terrestrial or extraterrestrial neutrino telescopes, which would assess probing Dark Energy models with this technique.

FewShot Upsampling for Protest Size Detection ; We propose a new task and dataset for a common problem in social science research upsampling coarse document labels to finegrained labels or spans. We pose the problem in a question answering format, with the answers providing the finegrained labels. We provide a benchmark dataset and baselines on a socially impactful task identifying the exact crowd size at protests and demonstrations in the United States given only orderofmagnitude information about protest attendance, a very small sample of finegrained examples, and Englishlanguage news text. We evaluate several baseline models, including zeroshot results from rulebased and questionanswering models, fewshot models finetuned on a small set of documents, and weakly supervised models using a larger set of coarselylabeled documents. We find that our rulebased model initially outperforms a zeroshot pretrained transformer language model but that further finetuning on a very small subset of 25 examples substantially improves outofsample performance. We also demonstrate a method for finetuning the transformer span on only the coarse labels that performs similarly to our rulebased approach. This work will contribute to social scientists' ability to generate data to understand the causes and successes of collective action.

SGPALM a Fast Physically Interpretable Tensor Graphical Model ; We propose a new graphical model inference procedure, called SGPALM, for learning conditional dependency structure of highdimensional tensorvariate data. Unlike most other tensor graphical models the proposed model is interpretable and computationally scalable to high dimension. Physical interpretability follows from the Sylvester generative SG model on which SGPALM is based the model is exact for any observation process that is a solution of a partial differential equation of Poisson type. Scalability follows from the fast proximal alternating linearized minimization PALM procedure that SGPALM uses during training. We establish that SGPALM converges linearly i.e., geometric convergence rate to a global optimum of its objective function. We demonstrate the scalability and accuracy of SGPALM for an important but challenging climate prediction problem spatiotemporal forecasting of solar flares from multimodal imaging data.

Inspecting the concept knowledge graph encoded by modern language models ; The field of natural language understanding has experienced exponential progress in the last few years, with impressive results in several tasks. This success has motivated researchers to study the underlying knowledge encoded by these models. Despite this, attempts to understand their semantic capabilities have not been successful, often leading to nonconclusive, or contradictory conclusions among different works. Via a probing classifier, we extract the underlying knowledge graph of nine of the most influential language models of the last years, including word embeddings, text generators, and context encoders. This probe is based on concept relatedness, grounded on WordNet. Our results reveal that all the models encode this knowledge, but suffer from several inaccuracies. Furthermore, we show that the different architectures and training strategies lead to different model biases. We conduct a systematic evaluation to discover specific factors that explain why some concepts are challenging. We hope our insights will motivate the development of models that capture concepts more precisely.

Spatiotemporal dynamics of biocrust and vegetation on sand dunes ; We propose a model to study the spatiotemporal dynamics of biocrust and vegetation cover on sand dunes. The model consists of two coupled partial nonlinear differential equations and includes diffusion and advection terms for modeling the dispersal of vegetation and biocrust and the effect of wind on them. In the absence of spatial variability, the model exhibits selfsustained relaxation oscillations and regimes of bistabilitythe first state is dominated by biocrust and the second by vegetation. We concentrate on the onedimensional dynamics of the model and show that the front that connects these two states propagates mainly due to the wind advection. In the oscillatory regime, the front propagation is complex. For low wind DP drift potential values, a series of spatially oscillatory domains develops as the front advances downwind. These domains form due to the oscillations of the spatially homogeneous states away from the front. However, for higher DP values, the dynamics is much more complex, becoming very sensitive to the initial conditions and exhibiting an irregular spatial pattern as small domains are created and annihilated during the front advance. Such irregular dynamics can be associated with the temporal variations of dune cover. In addition, similar behavior can be generated by other models that exhibit temporal oscillations and bistability.

Why Would I Trust Your Numbers On the Explainability of Expected Values in Soccer ; In recent years, many different approaches have been proposed to quantify the performances of soccer players. Since player performances are challenging to quantify directly due to the lowscoring nature of soccer, most approaches estimate the expected impact of the players' ontheball actions on the scoreline. While effective, these approaches are yet to be widely embraced by soccer practitioners. The soccer analytics community has primarily focused on improving the accuracy of the models, while the explainability of the produced metrics is often much more important to practitioners. To help bridge the gap between scientists and practitioners, we introduce an explainable Generalized Additive Model that estimates the expected value for shots. Unlike existing models, our model leverages features corresponding to widespread soccer concepts. To this end, we represent the locations of shots by fuzzily assigning the shots to designated zones on the pitch that practitioners are familiar with. Our experimental evaluation shows that our model is as accurate as existing models, while being easier to explain to soccer practitioners.

CorpusBased Paraphrase Detection Experiments and Review ; Paraphrase detection is important for a number of applications, including plagiarism detection, authorship attribution, question answering, text summarization, text mining in general, etc. In this paper, we give a performance overview of various types of corpusbased models, especially deep learning DL models, with the task of paraphrase detection. We report the results of eight models LSI, TFIDF, Word2Vec, Doc2Vec, GloVe, FastText, ELMO, and USE evaluated on three different public available corpora Microsoft Research Paraphrase Corpus, Clough and Stevenson and Webis Crowd Paraphrase Corpus 2011. Through a great number of experiments, we decided on the most appropriate approaches for text preprocessing hyperparameters, submodel selectionwhere they exist e.g., Skipgram vs. CBOW, distance measures, and semantic similarityparaphrase detection threshold. Our findings and those of other researchers who have used deep learning models show that DL models are very competitive with traditional stateoftheart approaches and have potential that should be further developed.

Tolerance in ModelDriven Engineering A Systematic Literature Review with ModelDriven Tool Support ; Managing models in a consistent manner is an important task in the field of ModelDriven Engineering MDE. Although restoring and maintaining consistency is desired in general, recent work has pointed out that always strictly enforcing consistency at any point of time is often not feasible in realworld scenarios, and sometimes even contrary to what a user expects from a trustworthy MDE tool. The challenge of tolerating inconsistencies has been discussed from different viewpoints within and outside the modelling community, but there exists no structured overview of existing and current work in this regard. In this paper, we provide such an overview to help join forces tackling the unresolved problems of tolerating inconsistencies in MDE. We follow the standard process of a Systematic Literature Review SLR to point out what tolerance means, how it relates to uncertainty, which examples for tolerant software systems have already been discussed, and which benefits and drawbacks tolerating inconsistencies entails. Furthermore, we propose a toolchain that helps conducting SLRs in computer science and also eases the reproduction of results. Relevant metadata of the collected sources is uniformly described in a textual modelling language and exported to the graph database Neo4j to query aggregated information.

EthicalAdvice Taker Do Language Models Understand Natural Language Interventions ; Is it possible to use natural language to intervene in a model's behavior and alter its prediction in a desired way We investigate the effectiveness of natural language interventions for readingcomprehension systems, studying this in the context of social stereotypes. Specifically, we propose a new language understanding task, Linguistic Ethical Interventions LEI, where the goal is to amend a questionanswering QA model's unethical behavior by communicating contextspecific principles of ethics and equity to it. To this end, we build upon recent methods for quantifying a system's social stereotypes, augmenting them with different kinds of ethical interventions and the desired model behavior under such interventions. Our zeroshot evaluation finds that even today's powerful neural language models are extremely poor ethicaladvice takers, that is, they respond surprisingly little to ethical interventions even though these interventions are stated as simple sentences. Fewshot learning improves model behavior but remains far from the desired outcome, especially when evaluated for various types of generalization. Our new task thus poses a novel language understanding challenge for the community.

Knowing More About Questions Can Help Improving Calibration in Question Answering ; We study calibration in question answering, estimating whether model correctly predicts answer for each question. Unlike prior work which mainly rely on the model's confidence score, our calibrator incorporates information about the input example e.g., question and the evidence context. Together with data augmentation via back translation, our simple approach achieves 510 gains in calibration accuracy on reading comprehension benchmarks. Furthermore, we present the first calibration study in the open retrieval setting, comparing the calibration accuracy of retrievalbased span prediction models and answer generation models. Here again, our approach shows consistent gains over calibrators relying on the model confidence. Our simple and efficient calibrator can be easily adapted to many tasks and model architectures, showing robust gains in all settings.

MPCBERT A PreTrained Language Model for MultiParty Conversation Understanding ; Recently, various neural models for multiparty conversation MPC have achieved impressive improvements on a variety of tasks such as addressee recognition, speaker identification and response prediction. However, these existing methods on MPC usually represent interlocutors and utterances individually and ignore the inherent complicated structure in MPC which may provide crucial interlocutor and utterance semantics and would enhance the conversation understanding process. To this end, we present MPCBERT, a pretrained model for MPC understanding that considers learning who says what to whom in a unified model with several elaborated selfsupervised tasks. Particularly, these tasks can be generally categorized into 1 interlocutor structure modeling including replyto utterance recognition, identical speaker searching and pointer consistency distinction, and 2 utterance semantics modeling including masked shared utterance restoration and shared node detection. We evaluate MPCBERT on three downstream tasks including addressee recognition, speaker identification and response selection. Experimental results show that MPCBERT outperforms previous methods by large margins and achieves new stateoftheart performance on all three downstream tasks at two benchmarks.

ProjectionBased Reduced Order Model for Simulations of Nonlinear Flows with Multiple Moving Objects ; This paper presents a reduced order approach for transient modeling of multiple moving objects in nonlinear crossflows. The Proper Orthogonal Decomposition method and the Galerkin projection are used to construct a reduced version of the nonlinear NavierStokes equations. The Galerkin projection implemented in OpenFOAM platform allows accurate impositions of arbitrary timedependent boundary conditions at the moving boundaries. A modelling technique based on moving domain and immersed boundary techniques is proposed to overcome the challenge of handling moving boundaries due to movements of the multiple objects. The model is demonstrated capable to capture the complex flow fields past one and two oscillating cylinders and the forces acting on the cylinders. Simulation time could be reduced by more than three orders for a small case on a fine mesh as compared to an existing method and could be more for large cases. In general, the simulation time of the reduced model is of order of seconds as compared to hours of the full order Computational Fluid Dynamics models.

Model Zoo A Growing Brain That Learns Continually ; This paper argues that continual learning methods can benefit by splitting the capacity of the learner across multiple models. We use statistical learning theory and experimental analysis to show how multiple tasks can interact with each other in a nontrivial fashion when a single model is trained on them. The generalization error on a particular task can improve when it is trained with synergistic tasks, but can also deteriorate when trained with competing tasks. This theory motivates our method named Model Zoo which, inspired from the boosting literature, grows an ensemble of small models, each of which is trained during one episode of continual learning. We demonstrate that Model Zoo obtains large gains in accuracy on a variety of continual learning benchmark problems. Code is available at httpsgithub.comgrasplyrlmodelzoocontinual.

Antiparticle detector models in QFT ; We analyze families of particle detector models that linearly couple to different kinds of fermionic and bosonic fields. We also study the response of these detectors to particle and antiparticle excitations of the field. We propose a simple linear complex scalar particle detector model that captures the fundamental features of fermionic field detectors similarly to how the UnruhDeWitt model captures the features of the light matter interaction. We also discuss why we do not need to limit ourselves to quadratic models commonly employed in past literature. Namely, we provide a physically motivated mechanism that restores U1 symmetry in these linear complex models.

Scaling Vision Transformers ; Attentionbased neural networks such as the Vision Transformer ViT have recently attained stateoftheart results on many computer vision benchmarks. Scale is a primary ingredient in attaining excellent results, therefore, understanding a model's scaling properties is a key to designing future generations effectively. While the laws for scaling Transformer language models have been studied, it is unknown how Vision Transformers scale. To address this, we scale ViT models and data, both up and down, and characterize the relationships between error rate, data, and compute. Along the way, we refine the architecture and training of ViT, reducing memory consumption and increasing accuracy of the resulting models. As a result, we successfully train a ViT model with two billion parameters, which attains a new stateoftheart on ImageNet of 90.45 top1 accuracy. The model also performs well for fewshot transfer, for example, reaching 84.86 top1 accuracy on ImageNet with only 10 examples per class.

Applying endogenous learning models in energy system optimization ; Conventional energy production based on fossil fuels causes emissions which contribute to global warming. Accurate energy system models are required for a costoptimal transition to a zeroemission energy system, an endeavor that requires an accurate modeling of cost reductions due to technological learning effects. In this review, we summarize common methodologies for modeling technological learning and associated cost reductions. The focus is on learning effects in hydrogen production technologies due to their importance in a lowcarbon energy system, as well as the application of endogenous learning in energy system models. Finally, we present an overview of the learning rates of relevant lowcarbon technologies required to model future energy systems.

Uncovering the History of Cosmic Inflation from Anomalies in Cosmic Microwave Background Spectra ; We propose an inflationary primordial feature model that can explain both the large and smallscale anomalies in the currently measured cosmic microwave background CMB anisotropy spectra, revealing a clip of adventurous history of the Universe during its primordial epoch. Although the model is currently statistically indistinguishable from the Standard Model, we show that future observations such as the Simons Observatory and LiteBIRD will complement each other in distinguishing the model differences due to their accurate Emode polarization measurements, and the PICO mission, if funded, can put stringent constraints on all characteristic properties. The model predicts a signal of classical primordial standard clock, which can also be used to distinguish the inflation and alternative scenarios in a modelindependent fashion.

Named Entity Normalization Model Using Edge Weight Updating Neural Network Assimilation Between KnowledgeDriven Graph and DataDriven Graph ; Discriminating the matched named entity pairs or identifying the entities' canonical forms are critical in text mining tasks. More precise named entity normalization in text mining will benefit other subsequent text analytic applications. We built the named entity normalization model with a novel Edge Weight Updating Neural Network. Our proposed model when tested on four different datasets achieved stateoftheart results. We, next, verify our model's performance on NCBI Disease, BC5CDR Disease, and BC5CDR Chemical databases, which are widely used named entity normalization datasets in the bioinformatics field. We also tested our model with our own financial named entity normalization dataset to validate the efficacy for more general applications. Using the constructed dataset, we differentiate named entity pairs. Our model achieved the highest named entity normalization performances in terms of various evaluation metrics.

Using heterogeneity in semisupervised transcription hypotheses to improve codeswitched speech recognition ; Modeling codeswitched speech is an important problem in automatic speech recognition ASR. Labeled codeswitched data are rare, so monolingual data are often used to model codeswitched speech. These monolingual data may be more closely matched to one of the languages in the codeswitch pair. We show that such asymmetry can bias prediction toward the bettermatched language and degrade overall model performance. To address this issue, we propose a semisupervised approach for codeswitched ASR. We consider the case of EnglishMandarin codeswitching, and the problem of using monolingual data to build bilingual transcription models'' for annotation of unlabeled codeswitched data. We first build multiple transcription models so that their individual predictions are variously biased toward either English or Mandarin. We then combine these biased transcriptions using confidencebased selection. This strategy generates a superior transcript for semisupervised training, and obtains a 19 relative improvement compared to a semisupervised system that relies on a transcription model built with only the bestmatched monolingual data.

CRFL Certifiably Robust Federated Learning against Backdoor Attacks ; Federated Learning FL as a distributed learning paradigm that aggregates information from diverse clients to train a shared global model, has demonstrated great success. However, malicious clients can perform poisoning attacks and model replacement to introduce backdoors into the trained global model. Although there have been intensive studies designing robust aggregation methods and empirical robust federated training protocols against backdoors, existing approaches lack robustness certification. This paper provides the first general framework, Certifiably Robust Federated Learning CRFL, to train certifiably robust FL models against backdoors. Our method exploits clipping and smoothing on model parameters to control the global model smoothness, which yields a samplewise robustness certification on backdoors with limited magnitude. Our certification also specifies the relation to federated learning parameters, such as poisoning ratio on instance level, number of attackers, and training iterations. Practically, we conduct comprehensive experiments across a range of federated datasets, and provide the first benchmark for certified robustness against backdoor attacks in federated learning. Our code is available at httpsgithub.comAIsecureCRFL.

Discrete Autoregressive Variational Attention Models for Text Modeling ; Variational autoencoders VAEs have been widely applied for text modeling. In practice, however, they are troubled by two challenges information underrepresentation and posterior collapse. The former arises as only the last hidden state of LSTM encoder is transformed into the latent space, which is generally insufficient to summarize the data. The latter is a longstanding problem during the training of VAEs as the optimization is trapped to a disastrous local optimum. In this paper, we propose Discrete Autoregressive Variational Attention Model DAVAM to address the challenges. Specifically, we introduce an autoregressive variational attention approach to enrich the latent space by effectively capturing the semantic dependency from the input. We further design discrete latent space for the variational attention and mathematically show that our model is free from posterior collapse. Extensive experiments on language modeling tasks demonstrate the superiority of DAVAM against several VAE counterparts.

Global Dynamics of a PredatorPrey Model with StateDependent MaturationDelay ; In this paper, a stage structured predatorprey model with general nonlinear type of functional response is established and analyzed. The statedependent time delay hereafter SDTD is the time taken from predator's birth to its maturity, formatted as a monotonical ly increasing, continuously differentiable and bounded function on the number of mature predator. The model is quite different from many previous models with SDTD, in the sense that the derivative of delay on the time is involved in the model. First, we have shown that for a large class of commonly used types of functional responses, including Holling types I, II and III, BeddingtonDeAngelistype hereafter BDtype, etc, the predator coexists with the prey permanently if and only if the predator's net reproduction number is larger than one unit; Secondly, we have discussed the local stability of the equilibria of the model; Finally, for the special case of BDtype functional response, we claim that if the system is permanent, that is, the derivative of SDTD on the state is small enough and the predator interference is large enough, then the coexistence equilibrium is globally asymptotically stable.

Scientific Language Models for Biomedical Knowledge Base Completion An Empirical Study ; Biomedical knowledge graphs KGs hold rich information on entities such as diseases, drugs, and genes. Predicting missing links in these graphs can boost many important applications, such as drug design and repurposing. Recent work has shown that generaldomain language models LMs can serve as soft KGs, and that they can be finetuned for the task of KG completion. In this work, we study scientific LMs for KG completion, exploring whether we can tap into their latent knowledge to enhance biomedical link prediction. We evaluate several domainspecific LMs, finetuning them on datasets centered on drugs and diseases that we represent as KGs and enrich with textual entity descriptions. We integrate the LMbased models with KG embedding models, using a router method that learns to assign each input example to either type of model and provides a substantial boost in performance. Finally, we demonstrate the advantage of LM models in the inductive setting with novel scientific entities. Our datasets and code are made publicly available.

Model of metameric locomotion in active directional filaments ; Locomotion in segmented animals, such as annelids and myriapods centipedes and millipedes, is generated by a coordinated movement known as metameric locomotion, which can be also implemented in robots designed to perform specific tasks. We introduce a theoretical model, based on an active directional motion of the head segment and a passive trailing of the rest of the body segments, in order to formalize and study the metameric locomotion. The model is specifically formulated as a steered OrnsteinUhlenbeck curvature process, preserving the continuity of the curvature along the whole body filament, and thus supersedes the simple active Brownian model, which would be inapplicable in this case. We obtain the probability density by analytically solving the FokkerPlanck equation pertinent to the model. We also calculate explicitly the correlators, such as the meansquare orientational fluctuations, the orientational correlation function and the meansquare separation between the head and tail segments, both analytically either via the FokkerPlanck equation or directly by either solving analytically or implementing it numerically from the Langevin equations. The analytical and numerical results coincide. Our theoretical model can help understand the locomotion of metameric animals and instruct the design of metameric robots.

Label prompt for multilabel text classification ; One of the key problems in multilabel text classification is how to take advantage of the correlation among labels. However, it is very challenging to directly model the correlations among labels in a complex and unknown label space. In this paper, we propose a Label Mask multilabel text classification model LMMTC, which is inspired by the idea of cloze questions of language model. LMMTC is able to capture implicit relationships among labels through the powerful ability of pretrain language models. On the basis, we assign a different token to each potential label, and randomly mask the token with a certain probability to build a label based Masked Language Model MLM. We train the MTC and MLM together, further improving the generalization ability of the model. A large number of experiments on multiple datasets demonstrate the effectiveness of our method.

Systemic Infinitesimal Overdispersion on Graphical Dynamic Models ; Stochastic models for collections of interacting populations have crucial roles in scientific fields such as epidemiology and ecology, yet the standard approach to extending an ordinary differential equation model to a Markov chain does not have sufficient flexibility in the meanvariance relationship to match data. To handle that, overdispersed Markov chains have previously been constructed using gamma white noise on the rates. We develop new approaches using Dirichlet noise to construct collections of independent or dependent noise processes. This permits the modeling of highfrequency variation in transition rates both within and between the populations under study. Our theory is developed in a general framework of timeinhomogeneous Markov processes equipped with a graphical structure, for which ecological and epidemiological models provide motivating examples. We demonstrate our approach on a widely analyzed measles dataset, adding Dirichlet noise to a classical SEIR SusceptibleExposedInfectedRecovered model. Our methodology shows improved statistical fit measured by loglikelihood and provides new insights into the dynamics of this biological system.

Multiphase field models for collective cell migration ; Confluent cell monolayers and epithelia tissues show remarkable patterns and correlations in structural arrangements and activelydriven collective flows. We simulate these properties using multiphase field models. The models are based on cell deformations and cellcell interactions and we investigate the influence of microscopic details to incorporate active forces on emerging phenomena. We compare four different approaches, one in which the activity is determined by a random orientation, one where the activity is related to the deformation of the cells and two models with subcellular details to resolve the mechanochemical interactions underlying cell migration. The models are compared with respect to generic features, such as solidtoliquid phase transitions, cell shape variability, emerging nematic properties, as well as vorticity correlations and flow patterns in large confluent monolayers and confinements. All results are compared with experimental data for a large variety of cell cultures. The appearing qualitative differences of the models show the importance of microscopic details and provide a route towards predictive simulations of patterns and correlations in cell colonies.

Bayesian inference for continuoustime hidden Markov models with an unknown number of states ; We consider the modeling of data generated by a latent continuoustime Markov jump process with a state space of finite but unknown dimensions. Typically in such models, the number of states has to be prespecified, and Bayesian inference for a fixed number of states has not been studied until recently. In addition, although approaches to address the problem for discretetime models have been developed, no method has been successfully implemented for the continuoustime case. We focus on reversible jump Markov chain Monte Carlo which allows the transdimensional move among different numbers of states in order to perform Bayesian inference for the unknown number of states. Specifically, we propose an efficient splitcombine move which can facilitate the exploration of the parameter space, and demonstrate that it can be implemented effectively at scale. Subsequently, we extend this algorithm to the context of modelbased clustering, allowing numbers of states and clusters both determined during the analysis. The model formulation, inference methodology, and associated algorithm are illustrated by simulation studies. Finally, We apply this method to real data from a Canadian healthcare system in Quebec.

On the Use of TwoWay Fixed Effects Models for Policy Evaluation During Pandemics ; In the context of the Covid19 pandemic, multiple studies rely on twoway fixed effects FE models to assess the impact of mitigation policies on health outcomes. Building on the SIRD model of disease transmission, I show that FE models tend to be misspecified for three reasons. First, despite misleading common trends in the pretreatment period, the parallel trends assumption generally does not hold. Second, heterogeneity in infection rates and infected populations across regions cannot be accounted for by regionspecific fixed effects, nor by conditioning on observable timevarying confounders. Third, epidemiological theory predicts heterogeneous treatment effects across regions and over time. Via simulations, I find that the bias resulting from model misspecification can be substantial, in magnitude and sometimes in sign. Overall, my results caution against the use of FE models for mitigation policy evaluation.

Scientific multiagent reinforcement learning for wallmodels of turbulent flows ; The predictive capabilities of turbulent flow simulations, critical for aerodynamic design and weather prediction, hinge on the choice of turbulence models. The abundance of data from experiments and simulations and the advent of machine learning have provided a boost to turbulence modeling efforts. However, simulations of turbulent flows remain hindered by the inability of heuristics and supervised learning to model the nearwall dynamics. We address this challenge by introducing scientific multiagent reinforcement learning SciMARL for the discovery of wall models for largeeddy simulations LES. In SciMARL, discretization points act also as cooperating agents that learn to supply the LES closure model. The agents selflearn using limited data and generalize to extreme Reynolds numbers and previously unseen geometries. The present simulations reduce by several orders of magnitude the computational cost over fullyresolved simulations while reproducing key flow quantities. We believe that SciMARL creates unprecedented capabilities for the simulation of turbulent flows.

Physicsconstrained deep neural network method for estimating parameters in a redox flow battery ; In this paper, we present a physicsconstrained deep neural network PCDNN method for parameter estimation in the zerodimensional 0D model of the vanadium redox flow battery VRFB. In this approach, we use deep neural networks DNNs to approximate the model parameters as functions of the operating conditions. This method allows the integration of the VRFB computational models as the physical constraints in the parameter learning process, leading to enhanced accuracy of parameter estimation and cell voltage prediction. Using an experimental dataset, we demonstrate that the PCDNN method can estimate model parameters for a range of operating conditions and improve the 0D model prediction of voltage compared to the 0D model prediction with constant operationconditionindependent parameters estimated with traditional inverse methods. We also demonstrate that the PCDNN approach has an improved generalization ability for estimating parameter values for operating conditions not used in the DNN training.

A Transformerbased Crossmodal Fusion Model with Adversarial Training for VQA Challenge 2021 ; In this paper, inspired by the successes of visionlanguage pretrained models and the benefits from training with adversarial attacks, we present a novel transformerbased crossmodal fusion modeling by incorporating the both notions for VQA challenge 2021. Specifically, the proposed model is on top of the architecture of VinVL model 19, and the adversarial training strategy 4 is applied to make the model robust and generalized. Moreover, two implementation tricks are also used in our system to obtain better results. The experiments demonstrate that the novel framework can achieve 76.72 on VQAv2 teststd set.

AudioCLIP Extending CLIP to Image, Text and Audio ; In the past, the rapidly evolving field of sound classification greatly benefited from the application of methods from other domains. Today, we observe the trend to fuse domainspecific tasks and approaches together, which provides the community with new outstanding models. In this work, we present an extension of the CLIP model that handles audio in addition to text and images. Our proposed model incorporates the ESResNeXt audiomodel into the CLIP framework using the AudioSet dataset. Such a combination enables the proposed model to perform bimodal and unimodal classification and querying, while keeping CLIP's ability to generalize to unseen datasets in a zeroshot inference fashion. AudioCLIP achieves new stateoftheart results in the Environmental Sound Classification ESC task, outperforming other approaches by reaching accuracies of 90.07 on the UrbanSound8K and 97.15 on the ESC50 datasets. Further it sets new baselines in the zeroshot ESCtask on the same datasets 68.78 and 69.40, respectively. Finally, we also assess the crossmodal querying performance of the proposed model as well as the influence of full and partial training on the results. For the sake of reproducibility, our code is published.

Graph space using both geometric and probabilistic structure to evaluate statistical graph models ; Statistical graph models aim at modeling graphs as random realization among a set of possible graphs. One issue is to evaluate whether or not a graph is likely to have been generated by one particular model. In this paper we introduce the edit distance expected value EDEV and compare it with other methods such as entropy and distance to the barycenter. We show that contrary to them, EDEV is able to distinguish between graphs that have a typical structure with respect to a model, and those that do not. Finally we introduce a statistical hypothesis testing methodology based on this distance to evaluate the relevance of a candidate model with respect to an observed graph.

Extensions to Multifidelity Monte Carlo Methods for Simulations of Chaotic Systems ; Multifidelity Monte Carlo methods often rely on a preprocessing phase consisting of standard Monte Carlo sampling to estimate correlation coefficients between models of different fidelity to determine the weights and number of samples for each level. For computationally intensive models, as are often encountered in simulations of chaotic systems, this upfront cost can be prohibitive. In this work, a correlation estimation procedure is developed for the case in which the highest and next highest fidelity models are generated via discretizing the same mathematical model using different resolution. The procedure uses discretization error estimates to estimate the required correlation coefficient without the need to sample the highest fidelity model, which can dramatically decrease the cost of the preprocessing phase. The method is extended to chaotic problems by using discretization error estimates that account for the statistical nature of common quantities of interest and the accompanying finite sampling errors that pollute estimates of such quantities of interest. The methodology is then demonstrated on a model problem based on the KuramotoSivashinsky equation.

Explanatory Pluralism in Explainable AI ; The increasingly widespread application of AI models motivates increased demand for explanations from a variety of stakeholders. However, this demand is ambiguous because there are many types of 'explanation' with different evaluative criteria. In the spirit of pluralism, I chart a taxonomy of types of explanation and the associated XAI methods that can address them. When we look to expose the inner mechanisms of AI models, we develop Diagnosticexplanations. When we seek to render model output understandable, we produce Explicationexplanations. When we wish to form stable generalizations of our models, we produce Expectationexplanations. Finally, when we want to justify the usage of a model, we produce Roleexplanations that situate models within their social context. The motivation for such a pluralistic view stems from a consideration of causes as manipulable relationships and the different types of explanations as identifying the relevant points in AI systems we can intervene upon to affect our desired changes. This paper reduces the ambiguity in use of the word 'explanation' in the field of XAI, allowing practitioners and stakeholders a useful template for avoiding equivocation and evaluating XAI methods and putative explanations.

Relational VAE A Continuous Latent Variable Model for Graph Structured Data ; Graph Networks GNs enable the fusion of prior knowledge and relational reasoning with flexible function approximations. In this work, a general GNbased model is proposed which takes full advantage of the relational modeling capabilities of GNs and extends these to probabilistic modeling with Variational Bayes VB. To that end, we combine complementary preexisting approaches on VB for graph data and propose an approach that relies on graphstructured latent and conditioning variables. It is demonstrated that Neural Processes can also be viewed through the lens of the proposed model. We show applications on the problem of structured probability density modeling for simulated and real wind farm monitoring data, as well as on the metalearning of simulated Gaussian Process data. We release the source code, along with the simulated datasets.

Modelling the Earth's magnetic field with magnetometer data from a Raspberry Pi on board the ISS ; We used data from a magnetometer on board the ISS to simulate the Earth's magnetic field. For that purpose, we generated two models from the data and compared them an offcentered dipole model and a centered multipolar model. We found out that measuring the magnetic field along only two orbits is not enough to provide a good estimate of the Earth's magnetic field, as they chart a too small portion of the Earth's surface. This was our original work within the Astro Pi Challenge, a project ran by ESA and the Raspberry Pi Foundation, and the magnetometer used was the one incorporated in the Raspberry Pi Sense Hat. Here, we improved by performing the same analysis using data from ten orbits and learned that the multipolar model already provides a good approximation to the Earth's magnetic field with this number of orbits. We also noticed a deviation of the data collected from what would be expected from the IGRF model, in both sets of data.

GALExtin An alternative online tool to determine the interstellar extinction in the Milky Way ; Estimates of interstellar extinction are essential in a broad range of astronomical research. In the last decades, several maps and models of the large scale interstellar extinction in the Galaxy have been published. However, these maps and models have been developed in different programming languages, with different user interfaces and inputoutput formats, which makes using and comparing results from these maps and models difficult. To address this issue, we have developed a tool called GALExtin urlhttpwww.galextin.org that estimates interstellar extinction based on both 3D modelsmaps and 2D maps available. The user only needs to provide a list with coordinates and distance and to choose a modelmap. GALExtin will then provide an output list with extinction estimates. It can be implemented in any other portal or model that requires interstellar extinction estimates. Here, a general overview of GALExtin is presented, along with its capabilities, validation, performance and some results.

Progress and opportunities in modelling environmentally assisted cracking ; Environmentally assisted cracking phenomena are widespread across the transport, defence, energy and construction sectors. However, predicting environmentally assisted fractures is a highly crossdisciplinary endeavour that requires resolving the multiple materialenvironment interactions taking place. In this manuscript, an overview is given of recent breakthroughs in the modelling of environmentally assisted cracking. The focus is on the opportunities created by two recent developments phase field and multiphysics modelling. The possibilities enabled by the confluence of phase field methods and electrochemomechanics modelling are discussed in the context of three environmental assisted cracking phenomena of particular engineering interest hydrogen embrittlement, localised corrosion and corrosion fatigue. Mechanical processes such as deformation and fracture can be coupled with chemical phenomena like local reactions, ionic transport and hydrogen uptake and diffusion. Moreover, these can be combined with the prediction of an evolving interface, such as a growing pit or a crack, as dictated by a phase field variable that evolves based on thermodynamics and local kinetics. Suitable for both microstructural and continuum length scales, this new generation of simulationbased, multiphysics phase field models can open new modelling horizons and enable Virtual Testing in harmful environments.

End extending models of set theory via power admissible covers ; Motivated by problems involving end extensions of models of set theory, we develop the rudiments of the power admissible cover construction over illfounded models of set theory, an extension of the machinery of admissible covers invented by Barwise as a versatile tool for generalizing modeltheoretic results about countable wellfounded models of set theory to countable illfounded ones. Our development of the power admissible machinery allows us to obtain new results concerning powersetpreserving end extensions and rank extensions of countable models of subsystems of mathsfZFC. The canonical extension mathsfKPmathcalP of KripkePlatek set theory mathsfKP plays a key role in our work; one of our results refines a theorem of Rathjen by showing that Sigma1mathcalPtextmathsfFoundation is provable in mathsfKPmathcalP without invoking the axiom of choice.

TwoStage Sector Rotation Methodology Using Machine Learning and Deep Learning Techniques ; Market indicators such as CPI and GDP have been widely used over decades to identify the stage of business cycles and also investment attractiveness of sectors given market conditions. In this paper, we propose a twostage methodology that consists of predicting ETF prices for each sector using market indicators and ranking sectors based on their predicted rate of returns. We initially start with choosing sector specific macroeconomic indicators and implement Recursive Feature Elimination algorithm to select the most important features for each sector. Using our prediction tool, we implement different Recurrent Neural Networks models to predict the future ETF prices for each sector. We then rank the sectors based on their predicted rate of returns. We select the best performing model by evaluating the annualized return, annualized Sharpe ratio, and Calmar ratio of the portfolios that includes the top four ranked sectors chosen by the model. We also test the robustness of the model performance with respect to lookback windows and look ahead windows. Our empirical results show that our methodology beats the equally weighted portfolio performance even in the long run. We also find that Echo State Networks exhibits an outstanding performance compared to other models yet it is faster to implement compared to other RNN models.

Auxiliary Class Based Multiple Choice Learning ; The merit of ensemble learning lies in having different outputs from many individual models on a single input, i.e., the diversity of the base models. The high quality of diversity can be achieved when each model is specialized to different subsets of the whole dataset. Moreover, when each model explicitly knows to which subsets it is specialized, more opportunities arise to improve diversity. In this paper, we propose an advanced ensemble method, called Auxiliary class based Multiple Choice Learning AMCL, to ultimately specialize each model under the framework of multiple choice learning MCL. The advancement of AMCL is originated from three novel techniques which control the framework from different directions 1 the concept of auxiliary class to provide more distinct information through the labels, 2 the strategy, named memorybased assignment, to determine the association between the inputs and the models, and 3 the feature fusion module to achieve generalized features. To demonstrate the performance of our method compared to all variants of MCL methods, we conduct extensive experiments on the image classification and segmentation tasks. Overall, the performance of AMCL exceeds all others in most of the public datasets trained with various networks as members of the ensembles.

Particle fluctuations and the failure of simple effective models for manybody localized phases ; We investigate and compare the particle number fluctuations in the putative manybody localized MBL phase of a spinless fermion model with potential disorder and nearestneighbor interactions with those in the noninteracting case Anderson localization and in effective models where only interaction terms diagonal in the Anderson basis are kept. We demonstrate that these types of simple effective models cannot account for the particle number fluctuations observed in the MBL phase of the microscopic model. This implies that assisted and pair hopping termsgenerated when transforming the microscopic Hamiltonian into the Anderson basiscannot be neglected. As a consequence, it appears questionable if the microscopic model possesses an exponential number of exactly conserved local charges. If such exactly conserved local charges do not exist, then particles are expected to ultimately delocalize for any finite disorder strength.

Asymptotic Dependence of In and OutDegrees in a Preferential Attachment Model with Reciprocity ; Reciprocity characterizes the information exchange between users in a network, and some empirical studies have revealed that social networks have a high proportion of reciprocal edges. Classical directed preferential attachment PA models, though generating scalefree networks, may give networks with low reciprocity. This points out one potential problem of fitting a classical PA model to a given network dataset with high reciprocity, and indicates alternative models need to be considered. We give one possible modification of the classical PA model by including another parameter which controls the probability of adding a reciprocated edge at each step. Asymptotic analyses suggest that large in and outdegrees become fully dependent in this modified model, as a result of the additional reciprocated edges.

Growing hyperbolic networks beyond two dimensions the generalised popularitysimilarity optimisation model ; Hyperbolic network models have gained considerable attention in recent years, mainly due to their capability of explaining many peculiar features of realworld networks. One of the most widely known models of this type is the popularitysimilarity optimisation PSO model, working in the native disk representation of the twodimensional hyperbolic space and generating networks with smallworld property, scalefree degree distribution, high clustering and strong community structure at the same time. With the motivation of better understanding hyperbolic random graphs, we hereby introduce the dPSO model, a generalisation of the PSO model to any arbitrary integer dimension d2. The analysis of the obtained networks shows that their major structural properties can be affected by the dimension of the underlying hyperbolic space in a nontrivial way. Our extended framework is not only interesting from a theoretical point of view but can also serve as a starting point for the generalisation of already existing twodimensional hyperbolic embedding techniques.

Uncertainty quantification for industrial design using dictionaries of reduced order models ; We consider the dictionarybased ROMnet Reduced Order Model framework T. Daniel, F. Casenave, N. Akkari, D. Ryckelynck, Model order reduction assisted by deep neural networks ROMnet, Advanced modeling and Simulation in Engineering Sciences 7 16, 2020 and summarize the underlying methodologies and their recent improvements. The main contribution of this work is the application of the complete workflow to a reallife industrial model of an elastoviscoplastic highpressure turbine blade subjected to thermal, centrifugal and pressure loadings, for the quantification of the uncertainty on dual quantities such as the accumulated plastic strain and the stress tensor, generated by the uncertainty on the temperature loading field. The dictionarybased ROMnet computes predictions of dual quantities of interest for 1008 Monte Carlo draws of the temperature loading field in 2 hours and 48 minutes, which corresponds to a speedup greater than 600 with respect to a reference parallel solver using domain decomposition, with a relative error in the order of 2. Another contribution of this work consists in the derivation of a metamodel to reconstruct the dual quantities of interest over the complete mesh from their values on the reduced integration points.

Relationaware Compositional Zeroshot Learning for AttributeObject Pair Recognition ; This paper proposes a novel model for recognizing images with composite attributeobject concepts, notably for composite concepts that are unseen during model training. We aim to explore the three key properties required by the task relationaware, consistent, and decoupled to learn rich and robust features for primitive concepts that compose attributeobject pairs. To this end, we propose the Blocked Message Passing Network BMPNet. The model consists of two modules. The concept module generates semantically meaningful features for primitive concepts, whereas the visual module extracts visual features for attributes and objects from input images. A message passing mechanism is used in the concept module to capture the relations between primitive concepts. Furthermore, to prevent the model from being biased towards seen composite concepts and reduce the entanglement between attributes and objects, we propose a blocking mechanism that equalizes the information available to the model for both seen and unseen concepts. Extensive experiments and ablation studies on two benchmarks show the efficacy of the proposed model.

Fair DecisionMaking for Food Inspections ; Data and algorithms are essential and complementary parts of a largescale decisionmaking process. However, their injudicious use can lead to unforeseen consequences, as has been observed by researchers and activists alike in the recent past. In this paper, we revisit the application of predictive models by the Chicago Department of Public Health to schedule restaurant inspections and prioritize the detection of critical food code violations. We perform the first analysis of the model's fairness to the population served by the restaurants in terms of average time to find a critical violation. We find that the model treats inspections unequally based on the sanitarian who conducted the inspection and that, in turn, there are geographic disparities in the benefits of the model. We examine four alternate methods of model training and two alternative ways of scheduling using the model and find that the latter generate more desirable results. The challenges from this application point to important directions for future work around fairness with collective entities rather than individuals, the use of critical violations as a proxy, and the disconnect between fair classification and fairness in the dynamic scheduling system.

Longitudinal network models and permutationuniform Markov chains ; We offer a general approach to modeling longitudinal network data, including exponential random graph models ERGMs, that vary according to certain discretetime Markov chains. We connect conditional and Markovian exponential families, permutationuniform Markov chains, various temporal ERGMs, and statistical considerations such as dyadic independence and exchangeability. By removing models' temporal dependence but not interpretability, our approach simplifies analysis of some network and autoregressive models from the literature, including closedform expressions for maximum likelihood estimators. We also introduce exponential random tmultigraph models, motivated by our result on replacing t observations of permutationuniform Markov chains of graphs with single observations of corresponding multigraphs.

Partisan Confidence Model for Group Polarization ; Models of opinion dynamics play a major role in various disciplines, including economics, political science, psychology, and social science, as they provide a framework for analysis and intervention. In spite of the numerous mathematical models of social learning proposed in the literature, only a few models have focused on or allow for the possibility of popular extreme beliefs' formation in a population. This paper closes this gap by introducing the Partisan Confidence PC model inspired by the foundations of the wellestablished sociopsychological theory of groupthink. The model hints at the existence of a tipping point, passing which the opinions of the individuals within a socalled social bubble are exaggerated towards an extreme position, no matter how the general population is united or divided. The results are also justified through numerical experiments, which provide new insights into the evolution of opinions and the groupthink phenomenon.

Extended shallowwater theories with thermodynamics and geometry ; Driven by growing momentum in twodimensional geophysical flow modeling, this paper introduces a general family of thermal rotating shallowwater models. The models are capable of accommodating thermodynamic processes, such as those acting in the ocean mixed layer, by allowing buoyancy to vary in horizontal position and time as well as with depth, in a polynomial fashion up to an arbitrary degree. Moreover, the models admit EulerPoincare variational formulations and possess LiePoisson Hamiltonian structure. Such a geometric property provides solid fundamental support to the theories described with consequences for numerical implementation and the construction of unresolved motion parametrizations. In particular, it is found that stratification halts the development of smallscale filament rollups recently observed in a popular model, which, having vertically homogeneous density, represents a special case of the models presented here.

The Impact of Mobility between Rural Areas and Forests on the Spread of Zika ; A mathematical model of Zika virus transmission incorporating human movement between rural areas and nearby forests is presented to investigate the role of human movement in the spread of Zika virus infections in human and mosquito populations. Proportions of both susceptible and infected humans living in rural areas are assumed to move to nearby forest areas. Direct, indirect and vertical transmission routes are incorporated for all populations. Mathematical analysis of the proposed model has been presented. The analysis starts with normalizing the proposed model. Positivity and boundedness of solutions to the normalized model have been then addressed. The basic reproduction number has been calculated using the next generation matrix method and its relation to the three routes of disease transmission has been presented. The sensitivity analysis of the basic reproduction number to all model parameters has been investigated. The analysis also includes existence and stability of disease free and endemic equilibrium points. Bifurcation analysis has been also carried out. Finally, numerical solutions to the normalized model have been obtained to confirm the theoretical results and to demonstrate the impact of human movement in the disease transmission in human and mosquito populations.

Separable Temporal Convolution plus Temporally Pooled Attention for Lightweight Highperformance Keyword Spotting ; Keyword spotting KWS on mobile devices generally requires a small memory footprint. However, most current models still maintain a large number of parameters in order to ensure good performance. In this paper, we propose a temporally pooled attention module which can capture global features better than the AveragePool. Besides, we design a separable temporal convolution network which leverages depthwise separable and temporal convolution to reduce the number of parameter and calculations. Finally, taking advantage of separable temporal convolution and temporally pooled attention, a efficient neural network STAttNet is designed for KWS system. We evaluate the models on the publicly available Google speech commands data sets V1. The number of parameters of proposed model 48K is 16 of stateoftheart TCResNet141.5 model 305K. The proposed model achieves a 96.6 accuracy, which is comparable to the TCResNet141.5 model 96.6.

This looks more like that Enhancing SelfExplaining Models by Prototypical Relevance Propagation ; Current machine learning models have shown high efficiency in solving a wide variety of realworld problems. However, their black box character poses a major challenge for the understanding and traceability of the underlying decisionmaking strategies. As a remedy, many posthoc explanation and selfexplanatory methods have been developed to interpret the models' behavior. These methods, in addition, enable the identification of artifacts that can be learned by the model as classrelevant features. In this work, we provide a detailed case study of the selfexplaining network, ProtoPNet, in the presence of a spectrum of artifacts. Accordingly, we identify the main drawbacks of ProtoPNet, especially, its coarse and spatially imprecise explanations. We address these limitations by introducing Prototypical Relevance Propagation PRP, a novel method for generating more precise modelaware explanations. Furthermore, in order to obtain a clean dataset, we propose to use multiview clustering strategies for segregating the artifact images using the PRP explanations, thereby suppressing the potential artifact learning in the models.

Geometric Models for Temporally Attributed Description Logics ; In the search for knowledge graph embeddings that could capture ontological knowledge, geometric models of existential rules have been recently introduced. It has been shown that convex geometric regions capture the socalled quasichained rules. Attributed description logics DL have been defined to bridge the gap between DL languages and knowledge graphs, whose facts often come with various kinds of annotations that may need to be taken into account for reasoning. In particular, temporally attributed DLs are enriched by specific attributes whose semantics allows for some temporal reasoning. Considering that geometric models and temporally attributed DLs are promising tools designed for knowledge graphs, this paper investigates their compatibility, focusing on the attributed version of a Horn dialect of the DLLite family. We first adapt the definition of geometric models to attributed DLs and show that every satisfiable ontology has a convex geometric model. Our second contribution is a study of the impact of temporal attributes. We show that a temporally attributed DL may not have a convex geometric model in general but we can recover geometric satisfiability by imposing some restrictions on the use of the temporal attributes.

On the Significance of Question Encoder Sequence Model in the OutofDistribution Performance in Visual Question Answering ; Generalizing beyond the experiences has a significant role in developing practical AI systems. It has been shown that current Visual Question Answering VQA models are overdependent on the languagepriors spurious correlations between questiontypes and their most frequent answers from the train set and pose poor performance on OutofDistribution OOD test sets. This conduct limits their generalizability and restricts them from being utilized in realworld situations. This paper shows that the sequence model architecture used in the questionencoder has a significant role in the generalizability of VQA models. To demonstrate this, we performed a detailed analysis of various existing RNNbased and Transformerbased questionencoders, and along, we proposed a novel Graph attention network GATbased questionencoder. Our study found that a better choice of sequence model in the questionencoder improves the generalizability of VQA models even without using any additional relatively complex biasmitigation approaches.

Enterprise Architecture Model Transformation Engine ; With increasing linkage within value chains, the IT systems of different companies are also being connected with each other. This enables the integration of services within the movement of Industry 4.0 in order to improve the quality and performance of the processes. Enterprise architecture models form the basis for this with a better buisness ITalignment. However, the heterogeneity of the modeling frameworks and description languages makes a concatenation considerably difficult, especially differences in syntax, semantic and relations. Therefore, this paper presents a transformation engine to convert enterprise architecture models between several languages. We developed the first generic translation approach that is free of specific metamodeling, which is flexible adaptable to arbitrary modeling languages. The transformation process is defined by various pattern matching techniques using a rulebased description language. It uses set theory and firstorder logic for an intuitive description as a basis. The concept is practical evaluated using an example in the area of a large German ITservice provider. Anyhow, the approach is applicable between a wide range of enterprise architecture frameworks.

CTAL Pretraining Crossmodal Transformer for AudioandLanguage Representations ; Existing audiolanguage taskspecific predictive approaches focus on building complicated latefusion mechanisms. However, these models are facing challenges of overfitting with limited labels and low model generalization abilities. In this paper, we present a Crossmodal Transformer for AudioandLanguage, i.e., CTAL, which aims to learn the intramodality and intermodality connections between audio and language through two proxy tasks on a large amount of audioandlanguage pairs masked language modeling and masked crossmodal acoustic modeling. After finetuning our pretrained model on multiple downstream audioandlanguage tasks, we observe significant improvements across various tasks, such as, emotion classification, sentiment analysis, and speaker verification. On this basis, we further propose a speciallydesigned fusion mechanism that can be used in finetuning phase, which allows our pretrained model to achieve better performance. Lastly, we demonstrate detailed ablation studies to prove that both our novel crossmodality fusion component and audiolanguage pretraining methods significantly contribute to the promising results.

A comment on Discrete time crystals rigidity, criticality, and realizations ; The Letter by N. Y. Yao et. al. 1,2 presents three models for realizing a manybody localized discrete timecrystal MBL DTC a shortranged model 1, its revised version 2, as well as a longrange model of a trapped ion experiment 1,3. We show that none of these realize an MBL DTC for the parameter ranges quoted in Refs. 1,2. The central phase diagrams in 1 therefore cannot be reproduced. The models show rapid decay of oscillations from generic initial states, in sharp contrast to the robust period doubling dynamics characteristic of an MBL DTC. Longlived oscillations from special initial states such as polarized states can be understood from the familiar lowtemperature physics of a static transverse field Ising model, rather than the nonequilibrium physics of an eigenstateordered MBL DTC. Our results on the longrange model also demonstrate, by extension, the absence of an MBL DTC in the trapped ion experiment of Ref. 3.

On Length Divergence Bias in Textual Matching Models ; Despite the remarkable success deep models have achieved in Textual Matching TM tasks, it still remains unclear whether they truly understand language or measure the semantic similarity of texts by exploiting statistical bias in datasets. In this work, we provide a new perspective to study this issue via the length divergence bias. We find the length divergence heuristic widely exists in prevalent TM datasets, providing direct cues for prediction. To determine whether TM models have adopted such heuristic, we introduce an adversarial evaluation scheme which invalidates the heuristic. In this adversarial setting, all TM models perform worse, indicating they have indeed adopted this heuristic. Through a welldesigned probing experiment, we empirically validate that the bias of TM models can be attributed in part to extracting the text length information during training. To alleviate the length divergence bias, we propose an adversarial training method. The results demonstrate we successfully improve the robustness and generalization ability of models at the same time.

An ExtraDimensional Model of Dark Matter ; We present a model for dark matter with extra spatial dimensions in which StandardModel SM fermions have localized wave functions. The underlying gauge group is Grm SM otimes rm U1z, and the dark matter particle is a SMsinglet Dirac fermion, chi, which is charged under the rm U1z gauge symmetry. We show that the conventional wisdom that the mass of a Dirac fermion is naturally at the ultraviolet cutoff scale does not hold in this model. We further demonstrate that this model yields a dark matter relic abundance in agreement with observation and discuss constraints from direct and indirect searches for dark matter. The dark matter particle interacts weakly with matter and has negligibly small selfinteractions. Very good fits to data from cosmological observations and experimental dark matter searches are obtained with mchi in the multiTeV range. A discussion is given of observational signatures and experimental tests of the model.

Selfexplaining variational posterior distributions for Gaussian Process models ; Bayesian methods have become a popular way to incorporate prior knowledge and a notion of uncertainty into machine learning models. At the same time, the complexity of modern machine learning makes it challenging to comprehend a model's reasoning process, let alone express specific prior assumptions in a rigorous manner. While primarily interested in the former issue, recent developments intransparent machine learning could also broaden the range of prior information that we can provide to complex Bayesian models. Inspired by the idea of selfexplaining models, we introduce a corresponding concept for variational GaussianProcesses. On the one hand, our contribution improves transparency for these types of models. More importantly though, our proposed selfexplaining variational posterior distribution allows to incorporate both general prior knowledge about a target function as a whole and prior knowledge about the contribution of individual features.

Modeling Massive Spatial Datasets Using a Conjugate Bayesian Linear Regression Framework ; Geographic Information Systems GIS and related technologies have generated substantial interest among statisticians with regard to scalable methodologies for analyzing large spatial datasets. A variety of scalable spatial process models have been proposed that can be easily embedded within a hierarchical modeling framework to carry out Bayesian inference. While the focus of statistical research has mostly been directed toward innovative and more complex model development, relatively limited attention has been accorded to approaches for easily implementable scalable hierarchical models for the practicing scientist or spatial analyst. This article discusses how pointreferenced spatial process models can be cast as a conjugate Bayesian linear regression that can rapidly deliver inference on spatial processes. The approach allows exact sampling directly avoids iterative algorithms such as Markov chain Monte Carlo from the joint posterior distribution of regression parameters, the latent process and the predictive random variables, and can be easily implemented on statistical programming environments such as R.

Emerging AI Security Threats for Autonomous Cars Case Studies ; Artificial Intelligence has made a significant contribution to autonomous vehicles, from object detection to path planning. However, AI models require a large amount of sensitive training data and are usually computationally intensive to build. The commercial value of such models motivates attackers to mount various attacks. Adversaries can launch model extraction attacks for monetization purposes or steppingstone towards other attacks like model evasion. In specific cases, it even results in destroying brand reputation, differentiation, and value proposition. In addition, IP laws and AIrelated legalities are still evolving and are not uniform across countries. We discuss model extraction attacks in detail with two usecases and a generic killchain that can compromise autonomous cars. It is essential to investigate strategies to manage and mitigate the risk of model theft.

DistantlySupervised Named Entity Recognition with NoiseRobust Learning and Language Model Augmented SelfTraining ; We study the problem of training named entity recognition NER models using only distantlylabeled data, which can be automatically obtained by matching entity mentions in the raw text with entity types in a knowledge base. The biggest challenge of distantlysupervised NER is that the distant supervision may induce incomplete and noisy labels, rendering the straightforward application of supervised learning ineffective. In this paper, we propose 1 a noiserobust learning scheme comprised of a new loss function and a noisy label removal step, for training NER models on distantlylabeled data, and 2 a selftraining method that uses contextualized augmentations created by pretrained language models to improve the generalization ability of the NER model. On three benchmark datasets, our method achieves superior performance, outperforming existing distantlysupervised NER models by significant margins.

PICARD Parsing Incrementally for Constrained AutoRegressive Decoding from Language Models ; Large pretrained language models for textual data have an unconstrained output space; at each decoding step, they can produce any of 10,000s of subword tokens. When finetuned to target constrained formal languages like SQL, these models often generate invalid code, rendering it unusable. We propose PICARD code and trained models available at httpsgithub.comElementAIpicard, a method for constraining autoregressive decoders of language models through incremental parsing. PICARD helps to find valid output sequences by rejecting inadmissible tokens at each decoding step. On the challenging Spider and CoSQL texttoSQL translation tasks, we show that PICARD transforms finetuned T5 models with passable performance into stateoftheart solutions.

College Student Retention Risk Analysis From Educational Database using MultiTask MultiModal Neural Fusion ; We develop a Multimodal Spatiotemporal Neural Fusion network for MultiTask Learning MSNFMTCL to predict 5 important students' retention risks future dropout, next semester dropout, type of dropout, duration of dropout and cause of dropout. First, we develop a general purpose multimodal neural fusion network model MSNF for learning students' academic information representation by fusing spatial and temporal unstructured advising notes with spatiotemporal structured data. MSNF combines a Bidirectional Encoder Representations from Transformers BERTbased document embedding framework to represent each advising note, LongShort Term Memory LSTM network to model temporal advising note embeddings, LSTM network to model students' temporal performance variables and students' static demographics altogether. The final fused representation from MSNF has been utilized on a MultiTask Cascade Learning MTCL model towards building MSNFMTCL for predicting 5 student retention risks. We evaluate MSNFMTCL on a large educational database consists of 36,445 college students over 18 years period of time that provides promising performances comparing with the nearest stateofart models. Additionally, we test the fairness of such model given the existence of biases.

Crosslingual Transfer of Monolingual Models ; Recent studies in zeroshot crosslingual learning using multilingual models have falsified the previous hypothesis that shared vocabulary and joint pretraining are the keys to crosslingual generalization. Inspired by this advancement, we introduce a crosslingual transfer method for monolingual models based on domain adaptation. We study the effects of such transfer from four different languages to English. Our experimental results on GLUE show that the transferred models outperform the native English model independently of the source language. After probing the English linguistic knowledge encoded in the representations before and after transfer, we find that semantic information is retained from the source language, while syntactic information is learned during transfer. Additionally, the results of evaluating the transferred models in source language tasks reveal that their performance in the source domain deteriorates after transfer.

Language Models are Fewshot Multilingual Learners ; Generalpurpose language models have demonstrated impressive capabilities, performing on par with stateoftheart approaches on a range of downstream natural language processing NLP tasks and benchmarks when inferring instructions from very few examples. Here, we evaluate the multilingual skills of the GPT and T5 models in conducting multiclass classification on nonEnglish languages without any parameter updates. We show that, given a few English examples as context, pretrained language models can predict not only English test samples but also nonEnglish ones. Finally, we find the incontext fewshot crosslingual prediction results of language models are significantly better than random prediction, and they are competitive compared to the existing stateoftheart crosslingual models.

Deep Spatiotemporal Sparse Decomposition for Trend Prediction and Anomaly Detection in Cardiac Electrical Conduction ; Electrical conduction among cardiac tissue is commonly modeled with partial differential equations, i.e., reactiondiffusion equation, where the reaction term describes cellular stimulation and diffusion term describes electrical propagation. Detecting and identifying of cardiac cells that produce abnormal electrical impulses in such nonlinear dynamic systems are important for efficient treatment and planning. To model the nonlinear dynamics, simulation has been widely used in both cardiac research and clinical study to investigate cardiac disease mechanisms and develop new treatment designs. However, existing cardiac models have a great level of complexity, and the simulation is often timeconsuming. We propose a deep spatiotemporal sparse decomposition DSTSD approach to bypass the timeconsuming cardiac partial differential equations with the deep spatiotemporal model and detect the time and location of the anomaly i.e., malfunctioning cardiac cells. This approach is validated from the data set generated from the CourtemancheRamirezNattel CRN model, which is widely used to model the propagation of the transmembrane potential across the cross neuron membrane. The proposed DSTSD achieved the best accuracy in terms of spatiotemporal mean trend prediction and anomaly detection.

Neural forecasting at scale ; We study the problem of efficiently scaling ensemblebased deep neural networks for multistep time series TS forecasting on a large set of time series. Current stateoftheart deep ensemble models have high memory and computational requirements, hampering their use to forecast millions of TS in practical scenarios. We propose NBEATSP, a global parallel variant of the NBEATS model designed to allow simultaneous training of multiple univariate TS forecasting models. Our model addresses the practical limitations of related models, reducing the training time by half and memory requirement by a factor of 5, while keeping the same level of accuracy in all TS forecasting settings. We have performed multiple experiments detailing the various ways to train our model and have obtained results that demonstrate its capacity to generalize in various forecasting conditions and setups.