Transfer Learning for Human & AI

For machine agents to successfully interact with humans in real-world settings, they will need to develop an understanding of human mental life . We present a benchmark consisting of a large dataset of procedurally generated 3D animations, AGENT (Action,Goal, Efficiency, coNstraint, uTility) We validateAGENT with human-ratings, propose an evaluation protocol emphasizing generalization, and compare two strong baselines built on Bayesian inverseplanning and Theory of Mind neural network .…

Constraint-based recommenders support users in the identification of items fitting their wishes and needs . Example domains are financialservices and electronic equipment . In this paper we show how divide-and-conquerbased (direct) diagnosis algorithms (no conflict detection is needed) can be used in constraint-based recommendation scenarios .…

Reward decomposition is a critical problem in centralized training withdecentralized execution~(CTDE) paradigm for multi-agent reinforcement learning . We propose a general meta-learning-based Mixing Network with MetaPolicy Gradient~(MNMPG) framework to distill the global hierarchy for delicatereward decomposition . Our method is generally applicable to theCTDE method using a monotonic mixing network .…

The aim is to synthesize anime-faces which are style-consistent with a given reference anime-face . We propose adouble-branch discriminator to learn both domain-specific distributions anddomain-shared distributions . The new generator architecture is proposed to simultaneously transfer color/ texture styles and transform local facial shapes into anime-like counterparts .…

This research proposes a new integrated framework for identifying safelanding locations and planning in-flight divert maneuvers . The proposed framework wasable to achieve 94.8% of successful landing in highly challenging landingsites where over 80$\%$ of the area around the initial target lading point ishazardous, by effectively updating the target landing site and feedback controlgain during descent .…

Mixed environments are notorious for the conflicts of selfish and social interests . Wepropose BAROCCO is an extension of these algorithms capable to balance individual and social incentives . The mechanism behind BAROccO is to train two distinctbut interwoven components that jointly affect each agent’s decisions .…

We present PADA: A Prompt-based AutoregressiveDomain Adaptation algorithm, based on the T5 model . Given a test example, PADA generates a unique prompt and then, conditioned on this prompt, labelsthe example with respect to the task . PADA strongly outperforms state-of-the-art approaches and additional strong baselines .…

A promising characteristic of Deep Reinforcement Learning (DRL) is itsability to learn optimal policy in an end-to-end manner without relying on feature engineering . Most approaches assume a fully observable statespace, i.e. fully observable Markov Decision Process (MDP) In real-worldrobotics, this assumption is unpractical, because of sensor issues such assensors’ capacity limitation and sensor noise .…

Logical Options Framework (LOF) learns policies that are satisfying, optimal, and composable . LOF efficiently learns policies thatsatisfy tasks by representing the task as an automaton and integrating it into learning and planning . We evaluate LOF on four tasks in discrete and continuous domains, including a 3D pick-and-place environment .…

Explanations are used in recommender systems for various reasons . Users have to be supported in making (high-quality) decisions more quickly . Explanation is designed in order to achieve specific goals such as increasing transparency of areendation or increasing a user’s trust in the recommender system .…

In generalized Nash equilibrium problems (GNEPs), a player’s set of feasibleactions is also impacted by the actions taken by other players in the game . We extend the framework of Ratliff et al. (2014) to find inverse optimization solutions for the class of GNEPs with joint constraints .…

We propose a novel CP-Net based composition approach to qualitatively select an optimal set of consumers for an IaaS provider . The provider’s and consumers’ qualitative preferences are captured using CP-Nets . A greedy-based and a heuristic-based consumer selection approaches are proposed that effectively reduce the search space of candidates in the composition .…

In machine learning, this approach is often referred to as transferlearning . In this work, we analyze surgical step recognition on four different laparoscopic surgeries: Cholecystectomy, Right Hemicolectomy, Sleeve Gastrectomy, and Appendectomy . We introduce a new architecture, the Time-Series Adaptation Network (TSAN), an architectureoptimized for transfer learning .…

Mitigating bias in machine learning systems requires refining our understanding of bias propagation pathways . A metric formeasuring bias amplification was introduced in the seminal work by Zhao et al. We introduceand analyze a new, decoupled metric for measuring bias amplification,$\text{BiasAmp}_{\rightarrow}$ (Directional Bias Amplification) We provide suggestions about its measurement by cautioning against predicting sensitive attributes, encouraging the use ofconfidence intervals due to fluctuations in the fairness of models across runs,and discussing the limitations of what this metric captures .…

Large datasets’ availability is catalyzing a rapid expansion of deep learning in general and computer vision in particular . In many domains, lack of training data may become an obstacle to the practical application of computer vision techniques . We introduce an image augmentation framework, which enablesus to enlarge the number of training samples while providing the data for such tasks as object detection, semantic segmentation, instancesegmentation, object counting, image denoising, and classification .…

We explore the use of residual networks and neural attention for argumentmining and in particular link prediction . The method we propose makes noassumptions on document or argument structure . We propose a residualarchitecture that exploits attention, multi-task learning, and makes use ofensemble .…

The MAP-Elites QD algorithm uses the learned latent space of the VAE as the search space for levels . The latent space captures the properties of the games whose levels we want to generate and blend, while MAP-elites searches this latent space to find a diverse set of levels optimizing a given objective such as playability .…

Communication is a important factor that enables agents to work cooperatively in multi-agent reinforcement learning (MARL) Most previous work uses continuous communication whose high representational capacity comes at the expense of interpretability . Allowing agents to learn their own discrete message protocol emerged from a variety of domains can increase theinterpretability for human designers and other agents .…

We propose amulti-task inverse reinforcement learning (IRL) algorithm, called \emph{inversetemporal difference learning} (ITD) that learns shared state features and per-agent successor features . We further show how to seamlesslyintegrate ITD with learning from online environment interactions, arriving at anovel algorithm for reinforcement learning with demonstrations, called $\Psi\Phi$-learning (pronounced `Sci-Fi’) We provide empirical evidence for the effectiveness of this method for improving RL, IRL,imitation, and few-shot transfer, and we derive worst-case bounds for its performance in zero-shot transfers to new tasks .…

Forensic speaker identification has been looking to shed light on the problem by quantifying how much a speech recording belongs to a particularperson in relation to a population . In this work, we explore the use of speechembeddings obtained by training a CNN using the triplet loss .…

Stuttering Events in Podcasts (SEP-28k) is a dataset containing over 28k clips labeled with five event types . The ability to automatically detect stuttering events in speech could helpspeech pathologists track an individual’s fluency over time or help improves speech recognition systems for people with atypical speech patterns .…

We propose an Open Source Software Project Evaluation and Selection TOol (OSS PESTO) Targeting OSS on Github, the largest OSS source code host, it facilitates theevaluation practice by enabling practitioners to compare candidates . It also allows in-time Github datacollection and customized evaluation that enriches its effectiveness and easeof use .…

In this paper, we explore using runtime verification to design safecyber-physical systems (CPS) We build upon the Simplex Architecture, where control authority may switch from an unverified and potentially unsafe advanced controller to a backup baseline controller in order to maintain system safety .…

Human values are what an individual or a society considers important in life . Ignoring these human values in software can pose difficulties or risks for all stakeholders . Agile methods are predominantly focused on delivering business values . Study suggests new and exclusive values-based artefacts(e.g.,…

Prior research has established that a small proportion of individuals dominate team communication during global software development . It is not known how these members’ contributions affect their teams’ knowledge diffusion process, or whether their personality profiles are responsible for their dominant presence .…

Bitcoin and Ethereum transactions present one of the largest real-world complex networks that are publicly available for study . Preference attachment continues to be a key factor in the evolution of both the Bitcoin and . Ethereum, the second most important cryptocurrency, continues to evolve, authors say .…

Moral foundations theory helps understand differences in morality across cultures . In this paper, we propose a model to predict moral foundations (MF) from social media trending topics . We also investigate whether differences inMF influence emotional traits .…

This work examines the discrete-time networked SIR(susceptible-infected-recovered) epidemic model . The infection and recovery parameters may be time-varying . We provide a sufficient condition for the SIR model to converge to the set of healthy states exponentially . We illustrate the results via simulations over northern Indiana, USA.…

Mistakes in binary conditions are a source of error in software systems . They happen when developers use, e.g., instead of = . Theseboundary mistakes are hard to find and impose manual, labor-intensive work for software developers . We train different models on approximately 1.6M examples with faults in different boundary conditions .…

Recent advances in neural-network based generative modeling of speech has great potential for speech coding . However, the performance of such models drops when the input is not clean speech, e.g., in the presence of noise, prevents its use in practical applications .…

In this paper, we develop a zonal-based flexible bus services (ZBFBS) by considering both passenger demands spatial (origin-destination or OD) and volume stochastic variations . Service requests are grouped by zonal OD pairsand number of passengers per request, and aggregated into demand categories .…

Traditional power system frequency dynamics are driven by Newtonian physics,where a synchronous generator (SG), the historical primary source of power,follows a deceleration frequency trajectory upon power imbalances . Subsequent to a disturbance, an SG will modifypre-converter, mechanical power as a function of frequency .…

The field of dynamical systems is being transformed by the mathematical tools emerging from modern computing and data science . Koopman spectral theory has emerged as a dominant perspective over the past decade . This linear representation of nonlinear dynamics has tremendous potential to enable the prediction,estimation, and control of non linear systems with standard textbook methods developed for linear systems .…

The heterogeneous use of library-referencing modes across Golang projects has caused numerous dependency management issues, incurring reference inconsistencies and even build failures . We reported 280 issues, among which 181 (64.6\%) issues have been confirmed, and 160 of them (88.4\%) have been fixed or areunder fixing .…

The decentralized solution of linear systems of equations arises as asubproblem in optimization over networks . Typical examples include the KKTsystem corresponding to equality constrained quadratic programs in distributedoptimization algorithms or in active set methods . This note presents a tailoredstructure-exploiting decentralized variant of the conjugate gradient method .…

The false data injection (FDI) attack is a crucial form of cyber-physical security problems facing cyber power systems . There is noresearch revealing the problem of FDI attacks facing voltage source converterbased high voltage direct current transmission (VSC-HVDC) systems . And finally, the modified IEEE-14 bus system is used to demonstrate that attackers are capable of disrupting the operation security of converter stations in VSC- HVDC systems by FDI attack strategies .…

False Data Injection (FDI) attacks against powersystem state estimation are agrowing concern for operators . The attack is based on a combinestopology learning technique, based only on branch and buspowerflows, and attacker-side pseudo-residual assessment to performstealthy FDIattacks with high confidence .…

LoopLearner addresses the problem of compiler instability by predicting which way of writing a loop will lead to efficient compiled code . Applying the transformations that our model deems most favorableprior to compilation yields an average speedup of 1.14x. When trying the top-3suggested transformations, the average speed up even increases to 1.29x.…

Linearized Reed–Solomon codes can outperform diversity codes based on cyclic division algebras at low SNRs . Simulation results show that the proposed codes outperform full diversity codes . We also provide sequential decoders for these codes and,more generally, space–time codes constructed from finite field codes .…

We consider the approximation of manifold-valued functions by embedding them into a higher dimensional space, applying a vector-valuedapproximation operator and projecting the resulting vector back to themanifold . We provide explicitconstants that depend on the reach of the embedded manifold .…

Structural Variations, SVs, in a genome can be linked to adisease or characteristic phenotype . SVs and SNPs in HLA loci would also serve as a medicaltransformation method for determining the success of an organ transplant for apatient, and predisposition to diseases apriori.…

In the Internet of Things (IoT) era, vehicles and other intelligent components in an intelligent transportation system (ITS) are connected, formingVehicular Networks . Game Theory (GT) can be used to model and analyze individual or group behaviors of communicatingentities in VNs .…

Inertia of the power grids is declining, which results in a faster drop insystem frequency in case of load-generation imbalance . Power grids with renewables are moresusceptible to FRO attacks and the inertia of synchronous generators plays acritical role in reducing the success of FRo attacks in power grids .…

Clustering based encoding scheme for structural variations and om-ics basedanalysis . Logistic regression to work best for death due to heart attack (MHHRTATT) phenotypic cause of death . Variant Id P1_M_061510_3_402_P at chromosome 3 &position 192063195 was found to be most highly associated to MHH RTATT.…

In this paper, we propose an algorithm to address the nonconvexmulti-group multicast beamforming problem . We formulate a convex relaxation of the problem as a semidefinite program in a real Hilbert space . We prove that the sequence of perturbations is bounded, so the algorithm is guaranteed to converge to afeasible point .…

The growing need for high-performance controllers in safety-critical applications like autonomous driving has been motivating the development offormal safety verification techniques . Tothis end, we provide a principled procedure to compute a safe and invariant setfor nonlinear dynamic bicycle models using efficient convex approximationtechniques .…

In federated learning system, parameter gradients are shared amongparticipants and the central modulator . The original data never leave the protected source domain . However, the gradient itself might carry enough information for precise inference of the original data . By reporting theirparameter gradients to the central server, client datasets are exposed toinference attacks from adversaries .…

In this paper, we design an algorithm thatsolves the Non-positive Circuit weight Problem (NCP) on this class ofparametric graphs . The proposed algorithm isbased on max-plus algebra and formal languages and runs faster than otherexisting approaches . It achieves strongly polynomial time complexity$\mathcal{O}(n^4)$ (where $n$ is the number of nodes in the graph) The proposed algorithms are based on max plus algebra, and run faster than existing approaches .…