Transfer Learning for Human & AI

Calculations of Fourier transform of a constant quantity over an area orvolume defined by polygons are often useful in modelingwave scattering, or in fourier-space filtering of real-space . If the system is discretized onto a regulararray, Fast Fourier techniques can speed up the resulting calculations .…

We study the convergence of the discontinuous Galerkin (DG) method applied to the advection-reaction equation on meshes with reentrant faces . On such meshes, the upwind numerical flux is not smooth, and so the numerical integration of the resulting face terms can only be expected to be first-order accurate .…

Matrix completion is a ubiquitous tool in machine learning and data analysis . It is possible to obtain low noise, high costobservations of individual entries or high noise, low cost observations of entire columns . We introduce a regression-based completion algorithm for this setting and experimentally verify the performance of our approach on both synthetic and real data sets .…

In the wake of COVID-19, every government huddles to find the best interventions that will reduce the number of infection cases while minimizing the economic impact . With many intervention policies available, how does one decide which policy is the best course of action?…

Hierarchical TaskNetwork (HTN) planning can provide effective means to solve such deployment problems . We present an HTN planning domain that models deployment problems as well as those in realistic Cloud environments . Cloud providers are facing a complex problem in configuring software applications ready for deployment .…

Current datasets on AVs predominantly fail to account for the latter, as they donot include an adequate number of events and associated details that involve pedestrian and vehicle interactions . We propose using virtual Reality (VR) data as a complementary resource to current datasets .…

Unmanned aerial vehicles (UAVs) are envisioned to support applications in next-generation wireless networks in both civil and military fields . Empowering UAVs networks intelligence by artificialintelligence (AI) especially machine learning (ML) techniques is inevitable andappealing to enable the aforementioned applications .…

In the domain of software engineering, our efforts as researchers to advise industry on which software practices might be applied most effectively are limited by our lack of evidence based information about the relationships between context and practice efficacy . We are in the exploratory stage of evolving amodel for context for situated software practices .…

Recent years have witnessed an upsurge of interest in the problem of anomalydetection on attributed networks . Hop counts based on both global and local contextual information can be served as indicators of anomaly . We propose a hop-countbased model (HCM) to detect anomalies by modeling both local and global contextual information .…

In this paper, we consider the Poisson interface problem $-\nabla^2u=f$ in $\Omega\setminus \Gamma$ with Dirichlet boundary condition . We propose a sixth order compact finitedifference scheme on uniform Cartesian grids . Our numerical experiments confirm the sixth accuracy order of the proposed compact finite differencescheme on uniform meshes .…

Federated learning (FL) is an emerging distributed machine learning paradigm that protects privacy and tackles the problem of isolated data islands . The experimental results show that CSAFL significantly improves test accuracy by more than +5% on four imbalanced federated datasets in a non-IID setting andcompare the baseline methods .…

Twitter can be viewed as a data source for Natural Language Processing (NLP)tasks . The continuously updating data streams on Twitter make it challenging totrace real-time topic evolution . We propose a framework formodeling fuzzy transitions of topic clusters . We extend our previous work oncrisp cluster transitions by incorporating fuzzy logic in order to enrich the underlying structures identified by the framework .…

The proposed design is based on a generalization of an autoencoder loss function that accommodates both message-wise and bit-wise UEP scenarios . In both scenarios, the generalized loss function can be adjusted using an associated weight vector totrade off error probabilities corresponding to different importance classes .…

The approach is based on mining Githistories and manipulating a degree of interest (DOI) model . It transforms sourcecode to revitalize feature log levels based on the “interestingness” of the surrounding code . The tool successfully analyzes99.26% of logging statements, increases log level distributions by $20% .…

The integrated sensing and communication (ISAC) has emerged as a key technology in future wireless systems . In this paper, we aim to provide a comprehensive survey for the current research progress on the fundamental limits of ISAC . We propose a systematic classification method for both traditional radio sensing (such as radar sensing) and ISAC so that they can be naturally incorporated into a unified framework .…

In this paper, we propose a single RF-chain multi-antenna full-duplex (FD)relay build with $b$-bit analog phase shifters . We compare the data rate of the proposed FD relaying system with the data rates of the same system but with the FD relay replaced by anideal intelligent reflecting surface (IRS) Our results show that the relayingsystem with 2-bit quantized analog phase .…

Reconfigurable intelligent surface (RIS) emerges as a promising technology for the next generation networks . Numerical results demonstrate that better coverage performance and higher energy efficiency can be achieved by the large-scale deployment of RISs . In addition, the tradeoff between the BS and RIS densities is investigated .…

We present a shared-memory algorithm to compute high-quality solutions to the balanced $k$-way hypergraph partitioning problem . High solution quality is achieved by parallelizing the core technique of the currently best sequential partitionerKaHyPar . This approach is made fast and scalable through intrusive algorithms and data structures that allow precise control of parallelism through atomic operations and fine-grainedlocking .…

We integrate a surrogate model, used for fitness value estimation, into a state-of-the-art P3-likevariant of the Gene-Pool Optimal Mixing Algorithm (GOMEA) and adapt the algorithm for solving non-binary combinatorial problems . Ensembling several models is a common Machine Learning technique to achieve better performance .…

Remote state monitoring over wireless is envisaged to play a pivotal role in enabling beyond 5G applications ranging from remote drone control to remotesurgery . The proposed split learning of a Koopman autoencoder can locally predict future states, and the prediction accuracy increases with the representation dimension and transmission power, says the authors .…

libEnsemble is a library aimed at particular scalability- and capability-stretching uses . It enables running concurrent instances of an application in dynamically allocated ensembles through an extensible Pythonlibrary . We highlight the structure, execution, and capabilities of the library on leading pre-exascale environments as well as advanced capabilities forexascale and beyond .…

This work proposes to use evolutionary computation as a pathway to allow anew perspective on the modeling of energy expenditure and recovery of an individual athlete during exercise . We revisit a theoretical concept called the”three component hydraulic model” which is designed to simulate metabolicsystems during exercise and which is able to address recently highlighted shortcomings of currently applied performance models .…

Few-shot object detection (FSOD) aims to detect objects using only few examples . Method incorporates a coarse-to-fine approach into the proposal based object detection framework . We propose to learn a lightweight matching network to measure the similarity between each spatialposition in the query image feature map and spatially-pooled class features, instead of the traditional object/nonobject classifier .…

The increasing share of renewables in the electrical energy generation mix comes along with an increasing uncertainty in power supply . We propose a metric-basedambiguity set with an additional constraint on dependence structure, usingcopula theory . We develop a conic reformulation which is kept generic such that it can be applied to any decision-making problem under uncertainty in powersystems .…

DeepDebug is a data-driven program repair approach which learns to detect and fixbugs in Java methods mined from real-world GitHub repositories . We framebug-patching as a sequence-to-sequence learning task consisting of two steps . We show that pretraining on source code programs improves thenumber of patches found by 33% as compared to supervised training from scratch .…

Stochastic Gradient Descent (SGD) has become the de facto way to train deepneural networks in distributed clusters . A critical factor in determining the training throughput and model accuracy is the choice of the parametersynchronization protocol . While Bulk Synchronous Parallel (BSP) often achieves better converged accuracy, the corresponding training throughput can be negatively impacted by stragglers .…

A key issue in citation content analysis seeks to understand citations based on the language used during the making of a citation . We find that within citation collocates, there is very low correlationbetween citation type and sentiment . We suggest that citations can be betterunderstood as claims-making devices where the citation type can be explained by understanding how two claims are being compared .…

Recently, Beierle and Leander found two new sporadic quadratic APNpermutations in dimension 9 . We show that a function EA-inequivalent to thosesporadic APN permutations can be obtained by just applyingEA transformations and inversion to the original permutations . We conjecture that $C_u$ is not APN if $m$ isgreater than $3$ and $u is not a 7th power .…

In this paper we present an extended variant of low rank parity check matrix(LRPC) codes that have received significant interests in recent years . It isshown that the extension indeed yields a superfamily of LRPC codes . The decoding method of the proposedcodes is also investigated .…

COMFORT is a new compiler fuzzing framework for detecting JS enginebugs and behaviors that deviate from the ECMAScript standard . In 200 hoursof automated concurrent testing runs, we discover bugs in all tested JS engines . We had identified 158 unique JS engine bugs, of which 129 have been verified, and 115 have already been fixed by the developers .…

The $k$-nearest neighbors algorithm is a simple and one of the most effective methods for outlier detection . So far, there have been no attemptsmade to develop a differentially private ($\epsilon$-DP) approach . Our approach yields nearly optimalperformance on real-world data with varying dimensions when compared to thenon-private versions of $k$.-NN…

We consider the extremal values of the stationary distribution of sparsedirected random graphs with given degree sequences . The graphs are generated by thedirected configuration model . In the same setting, we prove that these results extend to the PageRankscores of the random digraph, thus confirming a version of the so-calledpower-law hypothesis .…

Understanding the fundamental limits of robust supervised learning hasemerged as a problem of immense interest . In thispaper, we determine optimal lower bounds on the cross-entropy loss in the presence of test-time adversaries, along with the corresponding optimalclassification outputs . We propose and provide a proof of correctnessfor a bespoke algorithm to compute this lower bound efficiently .…

Storm water detention ponds are used to manage the discharge of rainfall runoff from urban areas to nearby streams . Detention ponds are currently designed based on static controls: the output flow of a pond is capped at a fixed value .…

Current Continuous Integration processes face significant intrinsic cybersecurity challenges . The idea is not only to solve and test formal or regulatory security requirements of source code but also to adhere to the same principles to the CI pipeline itself .…

A large-scale collective roboticlearning system can acquire a repertoire of behaviors simultaneously, sharing exploration, experience, and representations across tasks . In this framework new tasks can be continuously instantiated from previously learned tasks . We develop a scalable and intuitive framework for specifying new tasks through user-provided examples of desired outcomes .…

Joint attention – the ability to purposefully coordinate attention with another agent – is a critical component of human social cognition . In this paper, we ask whether joint attention can beuseful as a mechanism for improving multi-agent coordination and sociallearning .…

Graph neural networks (GNNs) have been widely used in deep learning ongraphs . They can learn effective node representations that achieve superior performances in graph analysis tasks such as node classification and nodeclustering . However, most methods ignore the heterogeneity in real-worldgraphs and fail to learn complex semantic representations .…

ZeRO-Infinity can fit models with tens and even hundreds of trillions of parameters for training on currentgeneration GPU clusters . It sustains over 25 petaflops on 512 NVIDIA V100 GPUs, while also demonstrating super linearscalability . An open source implementation is available through DeepSpeed, a deep learning optimization library that makes distributedtraining easy, efficient, and effective .…

Pre-trained language models (LM) have become go-to text representationencoders . However, dense encoders suffer in low resource situations . One cause liesin the readiness of the LM to expose its knowledge through dense representation, which we term Optimization Readiness . We present Condenser, a general pre-training architecture based on TransformerLMs, to improve dense optimization readiness.…

Millions of repetitive code snippets are submitted to code repositories everyday . To search from these large codebases using simple natural language queries would allow programmers to ideate, prototype, and develop easier and faster . We show that our modellearns the inherent relationship between the embedding spaces and furtherprobes into the scope of improvement by empirically analyzing the embeddeddingmethods.…

Pylot is a platform for autonomous vehicle (AV) research and development . It is built with the goal to allow researchers to study the effects of latency and accuracy of their models and algorithms on the end-to-end driving behavior of an AV .…

In this article, a reliable and efficient a posteriori error estimator ofresidual type is derived for a class of discontinuous Galerkin methods for thefrictional contact problem with reduced normal compliance . We further derive a priori error estimates in theenergy norm under the minimal regularity assumption on the exact solution .…

Recent advances in quantum computing have drawn considerable attention tobuilding realistic application for and using quantum computers . However,designing a suitable quantum circuit architecture requires expert knowledge . We propose aquantum architecture search framework with the power of deep reinforcementlearning (DRL) to address this challenge .…

In this paper, we tackle the problem of human-robot coordination in sequences of manipulation tasks . We first devise a hierarchical prediction approach by combining Inverse Reinforcement Learning and short-term motion prediction using a Recurrent Neural Network . In a secondstep, we propose a dynamic version of the TAMP algorithm Logic- GeometricProgramming (LGP) Our version of Dynamic LGP, replans periodically to handle the mismatch between the human motion prediction and the actual human behavior .…

Product quantization (PQ) is a popular approach for maximum inner productsearch (MIPS) It is widely used in ad-hoc retrieval . Recent studies proposedifferentiable PQ, where embedding and quantization modules can be trainedjointly . In this work, we proposeSearch-oriented Product Quantization (SoPQ), where a novel training objectiveMCL is formulated .…

Behavior Trees (BTs) provide a formal framework for comparing the expressivepower of Behavior Trees to other action selection architectures . This leads to an expressiveness hierarchy of control architectures, which includes BTs,Decision Trees (DTs), Teleo-reactive Programs (TRs) and Finite State Machines (FSMs) By distinguishing between BTs with auxiliary variables and thosewithout, we demonstrate the existence of a trade-off in BT design between readability and expressiveness .…

Robot-Assisted Therapy (RAT) has successfully been used to improve social skills in children with autism spectrum disorders (ASD) The DREAM project explores how to deliver effective RAT interventions to ASD children within asupervisedautonomy framework for controlling the robotic agent . This could reduce the burden on the clinicians delivering such interventions .…

In this paper, we present a task space-based local motion planner that incorporates collision avoidance and constraints on end-effector motion during the execution of a task . We show that it may be possible tocompute a path between two given task space poses by directly moving from thestart to the goal pose, even if there are potential collisions with obstacles .…