Transfer Learning for Human & AI

Each auto doc tool comes with its own unique representation of API information . In this paper, I use aninformation visualization analysis of auto docs to generate potential design principles for improving their usability . Developers use auto docs as areference by looking up relevant API primitives given partial information, or leads, about its name, type, or behavior .…

The ability of robots to grasp novel objects has industry applications ine-commerce order fulfillment and home service . We present a Thompson sampling algorithm that learns to grasp a given object with unknowngeometry using online experience . We find that seeding the policy with the Dex-Net prior allows it to more efficiently find robust grasps on these objects .…

En-route congestion causes delays in air traffic networks and will becomemore prominent as air traffic demand will continue to increase . The Multi-layer Air Traffic Network Delay (MATND) model is developed by a data-driven approach, taking aircrafttracking data and flight schedules as inputs to characterize a national airtraffic network .…

An image is worth a thousand words, conveying information that goes beyond the mere visual content therein . In this paper, we study the intent behind social media images with an aim to analyze how visual information can assist recognition of human intent .…

This chapter presents an approach to embed the input/state/output constraints into the trajectory design for differentially flat systems . Using the flatness property, the system’s inputs/states can be expressed as a combination of Bezier curved flat outputs and theirderivatives . By applying desired constraints to the latter, we find the feasible regions for the output Beziers controlpoints i.e.…

The optimal attack signal is a step function which switches its sign at the final stage, and the optimal attack location is the node with the largest impedance to the grid substation . Numerical examples on a European benchmark model verify thedeveloped results .…

Bandits for OnlineRapid Grasp Exploration Strategy (BORGES) leverages the structure of the problem to efficiently discover high performing grasps for each object stablepose . BORGES can be used to complement any general-purpose grasping algorithm with any grasp modality (parallel-jaw, suction, multi-fingered, etc) to learnpolicies for objects in which they exhibit persistent failures .…

We present an explicit and efficient algebraic construction of capacity-achieving list decodable codes with both constant alphabet and list sizes . We use algebraic-geometric (AG) codes with evaluation points restricted to a subfield . We further show how to explicitly construct such BTT evasive subspaces, based on the explicit subspace designs of Guruswamiand Kopparty (Combinatorica, 2016), and composition .…

We provide a novel connection between $\mathsf{StoqMA}$ and the distribution testing via reversible circuits . We prove that easy-witness$\maths$stoqma$ is contained in a sub-class of $MA$ . Our results make a step towardscollapsing the hierarchy, in which all classes are contained in $AM$ and $NP$ under derandomization assumptions .…

Model-free reinforcement learning (RL) is widely used to learn optimal policies for a variety of planning and control problems . Hamiltonian Q-Learning is a data efficient modification of the Q-learning approach . The proposed approach broadens the scope of RL algorithms for real-world applications, including classical control tasks and environmental monitoring, according to the proposed approach .…

A new self-supervised system for learning to detect novel and previously unseen categories of objects in images . The proposed systemreceives as input several unlabeled videos of scenes containing various objects . The frames of the videos are segmented into objects using depthinformation, and the segments are tracked along each video .…

Thispaper introduces an easily-tunable policy generation algorithm for autonomousdriving agents . The proposed algorithm balances diversity and driving skills byleveraging the representation and exploration abilities of deep reinforcementlearning via a distinct policy set selector . We present an algorithmutilizing intrinsic rewards to widen behavioral differences in the training.…

Reinforcement learning has shown great promise in robotics thanks to itsability to develop efficient robotic control procedures through self-training . In particular, reinforcement learning has been successfully applied to solving the reaching task with robotic arms . The Hindsight Experience Replay technique increases the average return of off-policy agents between7 and 9 folds when the target position is initialised randomly at the beginning of each episode .…

The vast literature on computational epidemiology offers solidgrounds for analytical studies and the definition of novel models aimed at bothpredictive and prescriptive scenario descriptions . UTLDR enables the design of compartmentalmodels incrementally and to simulate them over complex interaction networktopologies .…

WaDeNet is an end-to-end model for mobile speech processing systems . It embeds waveletdecomposition of the speech signal within the architecture . This allows the system to learn from spectral features in an end to-end manner, thus alleviating the need for feature extraction and successive modules that are currently present in speech processing .…

Edge intelligence has emerged as a promising paradigm to support mobile artificial intelligent(AI) applications at the network edge . We formulate a mixed-integer non-linear programming (MINLP) problem to minimize the total computation time and energy consumption of all users by jointly optimizing the service placement (i.e.…

We study approaches to robust model-based design of experiments in the context of maximum-likelihood estimation . The proposed methodology aims at problems where the experimentsare designed sequentially with a possibility of re-estimation in-between the experiments . The multi-stage formalism aids in identifying experiments that are better conducted in the early phase of experimentation, where parameterknowledge is poor .…

Deep artificial neural networks have been proposed as a model of primatevision . However, these networks are vulnerable to adversarial attacks, wherebyintroducing minimal noise can fool networks into misclassifying images . Primatevision is thought to be robust to such adversarial images .…

Edge computing has been recently introduced as a way to bring computational capabilities closer to end users of modern network-based services . But the problem of efficiently and fairly managing the system resources presents particular challenges due to the limited capacity of both edge nodes and wireless access networks .…

The Stewart Platform Robot for Interactive TabletopEngagement (SPRITE) is designed for use in socially assistiverobotics, a field focusing on non-contact social interaction to help people achieve goals relating to health, wellness, and education . We describe a series of design goals for a tabletop, socially assistive robot, including expressive movement, affective communication, a friendly, nonthreatening, and customizableappearance .…

We present a physics-constrained control-oriented deep learning method formodeling building thermal dynamics . The proposed method is based on thesystematic encoding of physics-based prior knowledge into a structuredrecurrent neural architecture . We demonstrate the proposed data-driven modeling approach’s effectiveness and physical interpretability on a dataset obtained from areal-world office building with 20 thermal zones .…

Humans are able to seamlessly visually imitate others, by inferring theirintentions and using past experience to achieve the same end goal . This paper investigates techniques which allow robots to partially bridge these domains gaps, using their past experience . A neural network is trained to mimic ground truth robot actions given context video from another agent, and must generalize to unseen taskinstances when prompted with new videos during test time .…

We present a novel method for handling uncertainty about the intentions of non-ego players in dynamic games . Equilibria in these games explicitly account for interaction among other agents in the environment, such as drivers and pedestrians . We leverage constraint asymmetries and feedback information patterns to incorporate the probabilities of hypotheses in a natural way .…

Hybrid refers to the inclusion of RTU andPMU measurement data, and the use of the LAV objective function enablesautomatic rejection of bad data . The proposed hybrid SE is formulated as a LP problem with guaranteedconvergence. To address efficiency, we further develop problem-specificheuristics for fast convergence.…

We develop data-driven algorithms for reachability analysis and control of systems with a priori unknown nonlinear dynamics . The resulting algorithms not only are suitable for settings with real-time requirements but also provide performance guarantees . They merge data from only asingle finite-horizon trajectory and, if available, various forms of side information .…

The structure theorem answers a question of De-Mossel-Neeman and of Ghazi-Kamath-Raghavendra . It is the first evidence for the optimality of the Frieze-Jerrum semidefiniteprogram for solving MAX-3-CUT . Without the assumption that each candidate has an equal chance of winning in (i) thePlurality is Stablest Conjecture is known to be false .…

This paper presents an online adaptive learning solution to optimalsynchronization control problem of heterogeneous multi-agent systems via anovel distributed policy iteration approach .…

This extended abstract reports on previous work of the CamFort project in which we developed an external units-of-measure type system for Fortran code . Our approach can guide the programmer in addingspecifications (type annotations) to existing code, with the aim of easing adoption on legacy code .…

We present an approach to continuous perception for robotic laundry tasks . Our assumption is that the visual prediction of a garment’s shapes and weights is possible via a neural network that learns the dynamic changes of garments from video sequences .…

The present distribution grids generally have limited sensing capabilities . Improved observability is a prerequisite for increasing the hosting capacity of distributed energy resources such as solar photovoltaics . The proposed solution incorporates voltage readings from neighboring CATV sensors, taking into account spatio-temporal aspects of the observations, and estimates single-phase voltage magnitudes at allnon-monitored buses using random forest .…

We formulate a new hardness assumption, the Strongish Planted CliqueHypothesis (SPCH) It postulates that any algorithm for planted clique must run in time $n^{\Omega(\log{n}$) We show that SPCH implies (nearly) tight inapproximability results for the following well-studied problems in terms of the parameter $k$ .…

FINO-Net is a novel multimodal sensor fusionbased deep neural network to detect and identify manipulation failures . The network combines RGB, depth and audioreadings to effectively detect and classify failures . Code and data are publicly available athttps://://://github.com/ardai/fino-net and the code and data is publicly available in code and public on the project’s GitHub page .…

The model is a recent variant of the P\’olya urn, introduced andstudied in arXiv:1906.10951 . It is characterized by a”local” reinforcement, i.e. a reinforcement mechanism mainly based on the mostrecent observations, and by a random persistent fluctuation of the predictivemean .…

Developing agile behaviors for legged robots remains a challenging problem . We formulate agile locomotion as a multi-stage learning problem in which a mentor guides the agent throughout the training . We evaluate our proposed learning system with a simulatedquadruped robot on a course consisting of randomly generated gaps and hurdles .…

Python package for modeling and studying the spread of diseases using an agent-based SEIR style epidemiological model with a focus on university campuses . Document explains the epidemiologicalmodel used in the package and gives examples highlighting the ways that the package can be used .…

Tactile skins have been shown to be useful for detecting collisions between a robotic arm and its environment . We propose a novel sensor design for use in covering existing multi-fingered robot hands . Assuch, we use a high density foam substrate to develop a scalable tactile skin that can be attached to the palm of a robotic hand .…

In recent years, edge computing has become a popular choice forlatency-sensitive applications like facial recognition and augmented reality because it is closer to the end users compared to the cloud . I-BOT is an interference-based orchestration scheme for latency-sensitive tasks on an Unmanaged Edge Platform (UEP) It minimizes the completion time ofapplications and is bandwidth efficient, authors say .…

We propose a method for offline learning of counterfactual predictions to address real world robotic reinforcement learning challenges . The proposedmethod encodes action-oriented visual observations as several “what if”questions learned offline from prior experience using reinforcement learning methods . We argue that the learned predictions form aneffective representation of the visual task, and guide the online explorationtowards high-potential success interactions (e.g.…

In high dimensional robotic system, the manifold of the valid configurationspace often has complex shape, especially under constraints such as end-effector orientation, static stability, and obstacles . We propose agenerative adversarial network approach to learn the distribution of validrobot configurations .…

Autonomous vehicles need precise knowledge on dynamic objects in their surroundings . The proposed approach shows a considerable performance increase, especially the number ofnon-continuous tracks is decreased significantly . The method is evaluated on simulations as well as on real scenarios.…

Localizing and tracking the pose of robotic grippers are necessary skills for manipulators . We propose a novelobservation model that conditions the tactile feedback on visual maps of the environment along with a motion model to recursively estimate the gripper’s location .…

We propose a novel transfer learning method for speech emotion recognition . With as low as 125 examples per emotion class, we were able to reach a higher accuracy than a strong baseline trained on 8 times more data . We alignacoustic pre-trained representations with semantic representations from theBERT model through an attention-based recurrent neural network .…

Sensor placement for contaminant detection in water distribution systems has become a topic of great interest aiming to secure a population’s water supply . Several approaches can be found in the literature ranging from the objective selected to optimize to the methodsimplemented to solve the optimization problem .…

Dynamic networks of concurrent pushdown systems (DCPS) are a theoretical model for multi-threaded recursive programs with shared global state . The (global) state reachability problem for DCPSis undecidable in general, but Atig et al. (2009) showed that it becomesdecidable, and is in 2EXPSPACE, when each thread is restricted to a fixed number of context switches .…

Optimisation algorithms are commonly compared on benchmarks to get insight into performance differences . However, it is not clear how closely benchmarksmatch the properties of real-world problems because these properties are largely unknown . This work investigates a questionnaire to enable the design of future benchmark problems that more closely resemble those found in the real world .…

The Internet of Medical Things (IoMT) offers an infrastructure made of smart medical equipment and software applications for health services . Patients can use their smart devices to create, store andshare their electronic health records (EHR) with medical personnel . However, unless the underlyingcombination within IoMT is secured, malicious users can intercept, modify and delete the sensitive EHR data of patients .…

Recurrent neural networks (RNNs) have shown significant improvements in recent years for speech enhancement . However, the model complexity andinference time cost of RNNs are much higher than deep feed-forward neuralnetworks (DNNs). Therefore, these limit the applications of speech enhancement. To make full use of the training data, we propose a full data learning method .…

Enhanced AutoNav (ENav), the baseline surface navigation software for NASA’sPerseverance rover, sorts a list of candidate paths for the rover to traverse, then uses the Approximate Clearance Evaluation (ACE) algorithm to evaluate whether the most highly ranked paths are safe .…

The task of playing musical instruments is largely unexplored because it involves the challenge of achieving sequential goals – melodies -that have a temporal dimension . In this paper, we address robotic musicianshipby introducing a temporal extension to goal-conditioned reinforcement learning:Time-dependent goals .…