Transfer Learning for Human & AI

Conventional traffic simulators usually employ a calibrated physical car-following model to describe vehicles’ behaviour . A fixed physical model tends to be lesseffective in a complicated environment given the non-stationary nature of traffic dynamics . In this paper, we formulate traffic simulation as an inversereinforcement learning problem, and propose a parameter sharing adversarialinverse reinforcement learning model for dynamics-robust simulation learning .…

We consider the problem of computing the Fr\’echet distance between twocurves for which the exact locations of the vertices are unknown . This problem was recently shown to be NP-hard in 2D, and it is unclear how to compute an optimalvertex placement at all .…

Empathetic response from the therapist is key to the success of clinicalpsychotherapy, especially motivational interviewing . Previous work oncomputational modelling of empathy in motivational interviewing has focused on the session-level assessment of therapist empathy . In this position paper, we propose anovel task of turn-level detection of client need for empathy .…

Deep learning has been used to demonstrate end-to-end neural network learningfor autonomous vehicle control from raw sensory input . We present an efficient and robust LiDAR-based end- to-end navigation framework . We evaluate our system on a full-scale vehicle anddemonstrate lane-stable as well as navigation capabilities .…

Unsupervised image registrational algorithms commonly employ intensity-based similarity measures as loss functions without any manual annotations . We propose anovel approach based on a low-rank representation, i.e., Regnet-LRR, to tackle the problem . We train the registration network to predict the dense deformationfields of noisy image pairs .…

The promise of socialnetworks like LinkedIn is that network friends enable easier access to critical resources such as legal and financial services, customers, and business partners . We find no positive effect of virtual network size and embeddedness, and small positive effects of location and diversity, between virtual and real-world networks .…

The Graph Augmented Multi-layer Perceptron (GA-MLP) model is an attractive alternative to Graph Neural Networks (GNNs) This is because it is resistant to the over-smoothing problem . The extended pdADMM-Q algorithm reduces communication cost by using quantization technique . Extensive experiments in six benchmarkdatasets demonstrate that the pdAdMM can lead to high speedup, and outperforms the existing state-of-the-art comparison methods .…

Noisy, intermediate-scale quantum (NISQ) computing devices offeropportunities to test the principles of quantum computing but are prone to errors arising from various sources of noise . Fluctuations in the noise itself lead to unstable devices that undermine the reproducibility of NISQ results .…

This paper presents a user-centered physical interface for collaborative mobile manipulators in industrial manufacturing and logistics applications . The proposed work builds on our earlier MOCA-MAN interface, through which a mobilemanipulator could be physically coupled to operators to assist them in daily activities .…

Modulo-wrapping receivers have attracted interest in several areas of digitalcommunications, including precoding and lattice coding . The asymptotic capacityand error performance of the modulo AWGN channel have been well established . However, due to underlying assumptions, these findings might not always be realistic in physical world applications, which are often dimension- or delay-limited .…

Modern machine learning algorithms rely on several design decisions to achieve strong performance . We combine the bandit-based HPO method Hyperband with the evolutionary search approach of Differential Evolution . DEHB is up to 1000x faster than random search . It is efficient in computational time, conceptually simple and easy to implement, positioning it well to become a new default HPO methods .…

This manuscripthighlights systems that are crucial for an emergency obstacle avoidancemaneuver (EOAM) It identifies the state-of-the-art for each of the related systems, while considering the nuances of traveling at highway speeds . Some of the primary EOAM-related systems/areas that are discussed in this review are: general path planning methods, system hierarchies, decision-making, trajectorygeneration, and trajectory-tracking control methods .…

Using a high degree of parallelism is essential to perform data assimilation efficiently . We examine approximations to the control variabletransform (CVT) technique when the latter is beneficial . The new strategyemploys a randomised singular value decomposition and retains the potential for parallelism in the time domain .…

The capability of individuals and teams can affect team performance and productivity . Measures associated with social aspects were observed to be dominant compared to technical and innovative aspects . The surveyedpractitioners suggested that an agile team member’s responsibility andquestioning skills significantly represent the member’s capability .…

The increasing concerns about data privacy and security drives the emergence of a new field of studying privacy-preserving machine learning from isolated data sources . We propose to protect the decision path by the efficient additivelyhomomorphic encryption method, which allows the disclosure of feature names and thus makes the federated decision trees interpretable .…

Geometric predicates are a basic ingredient to implement a vast range of algorithms in computational geometry . Modern implementations employ floatingpoint filtering techniques to combine efficiency and robustness . If the input to these predicates is an intermediate construction, its floating point representation may be affected by anapproximation error, and correctness is no longer guaranteed .…

Many technologies such as Deep Learning and tools like Word Embeddings havestarted to be investigated only recently . Many challenges remain open whenit comes to healthcare domain applications . To address these challenges, we propose the use of Deep Learning to identify sixteen morbidity types within textual descriptions of clinical records .…

The technology for Visual Odometry (VO) that estimates the position and orientation of the moving object through analyzing the image sequences captured by on-board cameras has been well investigated . This paper studies monocular VO from the perspective ofDeep Learning (DL) Unlike most current learning-based methods, our approach,called DeepAVO, is established on the intuition that features contributediscriminately to different motion patterns .…

Given a video captured from a first person perspective and recorded in afamiliar environment, can we recognize what the person is doing and identify where the action occurs in the 3D space? We address this challenging problem of jointly recognizing and localizing actions of a mobile user on a known 3D map from egocentric videos .…

The first generic self-stabilizing transformer for local problems in aconstrained bandwidth model is introduced . This transformer can be applied to awide class of locally checkable labeling (LCL) problems . The resulting algorithms are anonymous, size-uniform, and \emph{fullyadaptive} in the sense that their time complexity is bounded as a function of the number $k$ of nodes that suffered faults (possibly at different times) since the last legal configuration .…

Smart home environments are designed to provide services that help improve the quality of life for the occupant via a variety of sensors and actuators installed throughout the space . Many automated actions taken by a smart homeare governed by the output of an underlying activity recognition system .…

In this paper, assuming multi-antenna transmitter and receivers, we considermulticast beamformer design for the weighted max-min-fairness (WMMF) problem in a multi-stream multi-group communication setup . Unlike the single-streamscenario, the WMMF objective in this setup is not equivalent to maximizing theminimum weighted SINR due to the summation over the rates of multiple streams .…

Monte Carlo filtering objectives (MCFOs) extend the choices of likelihood estimators beyond Sequential Monte Carlo instate-of-the-art objectives . We demonstrate that the proposed MCFOs and gradient estimations lead to efficient and stable model learning, and learned generative models are more sample efficient on variouskinds of time series data .…

We consider the problem of finding an approximate solution to $ell_1$regression while only observing a small number of labels . We show that sampling from $X$ according to its Lewis weights and outputting theempirical minimizer succeeds with probability $1-\delta$ for $m O .…

We present an implicit Split-Step explicit Euler type Method (dubbed SSM) for the simulation of McKean-Vlasov Stochastic Differential Equations (MV-SDEs) The scheme is designed to leverage the structure induced by the interacting particle approximationsystem . The scheme attains the classical one-half root mean square error(rMSE) convergence rate in stepsize and closes the gap left by [18] regarding efficient implicit methods and theirconvergence rate for this class of SDEs .…

DeepDebug is an approach to automated debugging using large, pretrained transformers . We train a bug-creation model on reversed commit data for the purpose of generating synthetic bugs . We increase the total number of fixes found by over 50% on the QuixBugs benchmark, while decreasing the false positive rate from 35% to 5% .…

Anaphora and ellipses are two common phenomena in dialogues . Without resolving referring expressions and information omission, dialogue systems may fail to generate consistent and coherent responses . In this work, we propose a novel joint learning framework of modeling coreference resolutionand query rewriting for complex, multi-turn dialogue understanding .…

Event detection (ED) aims at detecting event trigger words in sentences and classifying them into specific event types . In real-world applications, ED typically does not have sufficient labelled data, thus can be formulated as afew-shot learning problem . We propose a novel knowledge-based few-shot event detection method which uses a definition-based encoder to introduce external event knowledge asthe knowledge prior of event types.…

A shape’s Heesch number is the number of layers of copies of the shape that can be placed around it without gaps or overlaps . Experimentation and searching have turned up examples of shapes with finite Heeschnumbers up to six, but nothing higher .…

Dependency parsing is a crucial step towards deep language understanding . Left-to-right and top-down transition-based algorithms that rely on Pointer Networks are among the most accurate approaches to performing dependency parsing . We develop a bottom-up-oriented HierarchicalPointer Network for the left-to the-right parser .…

Humans can flexibly leverage the compositional structure of the visual world, understanding new concepts as combinations of existing concepts . People can make meaningful compositional generalizations from just afew examples in a variety of scenarios . Bayesian programinduction model that provides a close fit to the behavioral data, says the authors of a new paper on how people learn different types of visualcompositions, using abstract visual forms with rich relational structure .…

Segtran is a novel Squeeze-and-Expansion transformer . The core of the new framework is a new positional encoding scheme for transformers, imposing a continuityinductive bias for images . Compared withrepresentative existing methods, Seg Tran consistently achieved the highestsegmentation accuracy, and exhibited good cross-domain generalizationcapabilities .…

Energy efficiency and energy conservation are one of the most crucial constraints for meeting the 20MW power envelope desired for exascale systems . We present a tuning plugin for thePeriscope Tuning Framework which integrates fine-grained autotuning at theregion level with DVFS and uncore frequency scaling (UFS) The tuning is based on a feed-forward neural network which is formulated using PerformanceMonitoring Counters (PMC) supported by x86 systems and trained using benchmarks .…

Internet banking (IB) continues to face security concerns arising from illegal access to users accounts . Use of personal identification numbers (PIN) as a single authentication method for IB users is prone to insecurities such asphishing, hacking and shoulder surfing .…

Channel estimation for hybrid Multiple Input Multiple Output (MIMO) systems at Millimeter-Waves (mmW/sub-THz) is a fundamental, despite challenging,prerequisite for an efficient design of hybrid MIMO precoding/combining . Mostworks propose sequential search algorithms, e.g., Compressive Sensing (CS), that are most suited to static channels .…

When data is of an extraordinarily large size or stored indifferent locations, the distributed nearest neighbor (NN) classifier is an attractive tool for classification . We propose a novel distributed adaptive NN classifier for which the number of nearest neighbors is a tuning parameter .…

Simulation speed depends on code structures, hence it is crucial how to build a fast algorithm . We solve the Allen-Cahn equation by an explicit finitedifference method, so it requires grid calculations implemented by manyfor-loops in the simulation code . We propose a modelarchitecture containing a pad and a convolution operation .…

Existing multi-camera solutions for automatic scorekeeping in steel-tip darts are very expensive and thus inaccessible to most players . We present a new approach to keypointdetection and apply it to predict dart scores from a single image taken fromany camera angle .…

Conversational agents (CAs) represent an emerging research field in healthinformation systems . There are great potentials in empowering patients with timely information and natural language interfaces . This paper provides practitioners with a blueprint for designing CAs in diabetes care with concrete design guidelines that can be extended into other healthcare domains .…

In recent years, abstractive text summarization with multimodal inputs has started drawing attention due to its ability to accumulate information from different source modalities . Existing methods use short videos as the visual modality and short summary asthe ground-truth, therefore, perform poorly on lengthy videos and longground-truth summary .…

The method is built upon a pyramidal convolutionalneural network architecture . It uses time recurrence and geometric constraints imposed by motion to produce pixel-wise depth maps . The code of our method is publicly available on GitHub . We analyse the performance of our approach on Mid-Air, a public drone dataset featuring synthetic dronetrajectories recorded in a wide variety of unstructured outdoor environments .…

Deep generative models of 3D shapes have received a great deal of research interest . Yet, almost all of them generate discrete shape representations, such as voxels, point clouds, and polygon meshes . We present the first 3D generative model for a drastically different shape representation — describing a shape as a sequence of computer-aided design (CAD) operations .…

ThundeRiNG is a resource-efficient and high-throughput system for generating multiple independent sequences of random numbers (MISRN) on FPGAs . It only consumes a constant number of DSPs (less than 1\% of the FPGA capacity) for generating any number of sequences, and achieves athroughput of 655 billion random numbers per second .…

We consider a class of variable effort human annotation tasks in which thenumber of labels required per item can greatly vary . On an image bounding-box task with crowdsourced annotators, we show that annotator accuracy and recallconsistently drop as effort increases .…

There are still many challenges in downlink channel estimation owing to inaccuratemodeling and high pilot overhead in practical systems . In this paper, we propose a channel state information (CSI) acquisition scheme for downlinkmassive MIMO-OTFS in presence of the fractional Doppler .…

The number of constraints depends on the size of the state and action spaces, which makes the problem intractable in large or continuous environments . A deepparameterized augment Lagrangian method is proposed . Preliminary experiments illustrate that our method iscompetitive to other state-of-the-art algorithms .…

The satisfiability checking problem for these extensions is undecidable when both the future U and the past S modalities are used . In a classical result, thesatisfiability checking for MITL[U,S] isshown to be decidable with EXPSPACE complete complexity . We propose a generalization of non punctuality called \emph{non adjacency} forTPTL[U] and focus on its 1-variable fragment .…

The mapped bases or Fake Nodes Approach (FNA), introduced in [10], allows to change the set of nodes without the need of resampling the function . However, the originally proposed S-Gibbs map suffers of a subtle instability when the interpolant is constructed at equidistant nodes, due to the Runge’s phenomenon .…

Recommender systems are gaining increasing and critical impacts on human and society since a growing number of users use them for information seeking and decision making . Just like users have personalized preferences on items, users’ demands for fairness are also personalized in many scenarios .…

With rapid development in artificial intelligence, social computing has evolved beyond social informatics toward the birth of social intelligencesystems . This paper takes initiatives to propose a social behaviourunderstanding framework with the use of deep neural networks for social andbehavioural analysis .…