Transfer Learning for Human & AI

The study also considers the recently proposed conformer, and more advanced self-attention based language models . Their combination and decoding reaches a new record on Switchboard-300 and 10.0% WER on SWB and CHM parts of Hub5’00 with very simple LSTMmodels. Overall, the conformer showssimilar performance to the L STM; nevertheless, their combination and decode with an improved LM reaches new record .…

CMA-Net consists of two novel mutual attention modules aiming at fusing the high level features from the modalities of all-in-focus and depth . It can run at a speed of 53 fps, thus being much faster than the state-of-the-art multi-modal SOD methods .…

Information retrieval (IR) systems traditionally aim to maximize metrics built on rankings, such as precision or NDCG . However, thenon-differentiability of the ranking operation prevents direct optimization of such metrics in state-of-the-art neural IR models . To address this shortcoming, we propose SmoothI, a smooth approximation of rank indicators that serves as abasic building block to devise differentiable approximations of IR metrics .…

Leader-boards like SuperGLUE are seen as important incentives for activedevelopment of NLP . We show that its test datasets are vulnerable to shallowheuristics . Often approaches based on simple rules outperform or come close to the results of the notorious pre-trained language models like GPT-3 or BERT .…

Recent studies have shown that providing explanations alongsiderecommendations increases trust and perceived quality . M&Ms-VAE is a novel autoencoder forrecommendation and explanation that is based on multimodal modelingassumptions . We train the model under a weak supervision scheme to simulate both fully and partially observed variables .…

The simplicity of the visual servoing approach makes it an attractive option for tasks dealing with vision-based control of robots in many real-world applications . Attaining precise alignment for unseen environments pose a challenge to existing visual servoational approaches . The recent data-driven approaches face issues when generalizing to novel environments .…

This article describes the goals and objectives of developing a software complex for planning the treatment of dentoalveolar anomalies . The architecture of the software complex as interacting components for treatment planning, as well as the principle of using algorithms using convolutional neural networks .…

This paper investigates the distributed consensus tracking control problem for general linear multi-agent systems (MASs) with external disturbances andheterogeneous time-varying input and communication delays . An extended LMI is proposed which, in conjunction with the rest of the LMIs, results in a solution with a larger upper bound on delays than what would befeasible without it .…

Joint radar and communication (RadCom) systems have attracted increased attention in recent years . Joint RadCom system is designed which marriesthe capabilities of a Multiple-Input Multiple-Output (MIMO) radar withRate-Splitting Multiple Access (RSMA) RSMA providesthe RadCom with more robustness, flexibility and user rate fairness compared to the baseline joint RadCom System based on Space Division Multiple Access(SDMA) System is designed in the presence of partial CSIT to maximize the Average Weighted Sum-Rate (AWSR) under QoS rate constraints and minimize the RadCom Beampattern Squared Error (BSE) against anideal MIMO radar beamp attern .…

The less abstract an emoji is, the more likely it is toundergo semantic change . We identify five patterns in emoji semantic development and analyse select emoji in more detail . We make our data publicly available alongwith a web-based interface that anyone can use to explore semantic change inemoji.…

Detecting pedestrians is a crucial task in autonomous driving systems . Relying solely on RGB cameras may not be enough to recognize roadenvironments in situations where cameras cannot capture scenes properly . Some approaches aim to compensate for these limitations by combining RGB cameras with TOF sensors, such as LIDARs .…

Technology-assisted review (TAR) refers to iterative active learning workflows for document review in high recall retrieval (HRR) tasks . We find that the pre-trained BERT model reduces review volume by 30% in TAR workflows simulated on RCV1-v2 newswire collection . In contrast, linear models outperform BERT for simulated legal discovery topics on Jeb Bush e-mail collection .…

In computational PDE-based inverse problems, a finite amount of data is collected to infer unknown parameters in the PDE . The best, and often infeasible, scenario is when the full input-to-output (ItO) map is available: This is the typical setting in many theoretical inverse problems .…

Viko is anadaptive gecko gripper that utilizes vision-based tactile sensors to monitor contact state . The sensor provides high-resolution real-time measurements of contact area and shear force . Viko integrated gecko-inspired adhesives into the sensor surface withoutimpeding its adaptiveness and performance .…

Reachability analysis of nonlinear dynamical systems is a challenging andcomputationally expensive task . Computing the reachable states for linear systems, in contrast, can often be done efficiently in high dimensions . The Koopman operator links the behaviors of a nonlinear system to a linear system embedded in a higher dimensional space, with an additional set of so-calledobservable variables .…

Weakly supervised segmentation requires assigning a label to every pixel based on training instances with partial annotations such as image-level tags,object bounding boxes, labeled points and scribbles . We propose 4 types of contrastiverelationships between pixels and segments in the feature space, capturinglow-level image similarity, semantic annotation, co-occurrence, and featureaffinity .…

This paper presents a robust controller using a Linear Parameter Varying(LPV) model of the lane-keeping system with parameter reduction . Multiple varying parameters lead to a high number of scheduling variables and cause massive computational complexity . We designed the LPV robust feedback controller using the reducedmodel solving a set of Linear Matrix Inequality (LMI) The effectiveness of the proposed system is validated with full vehicle dynamics from CarSim on aninterchange road .…

Trust between team members is an essential requirement for any successful cooperation . Engendering and maintaining the fellow team members’ trust becomes a central responsibility for any member trying to not only successfully participate in the task but to ensure the team achieves its goals .…

Standard losses for training deep segmentation networks could be seen as individual classifications of pixels, instead of supervising the global shape of the predicted segmentations . While effective, they require exact knowledge of the label of each pixel in an image .…

Signal processing and communication communities have witnessed the rise of exciting communication technologies in recent years . We discuss the recent developments in the field and put forward promising candidates for future research and development . We also envision an ultimate RIS architecture, which is able to adjust its operation modes dynamically, and introduce the new concept of PHY slicing over RISs towards 6G wireless networks.…

We introduce generalized spatially coupled parallel concatenated codes (GSC-PCCs) as a class of turbo-like turbo-style codes . We show that the proposed codes have some niceproperties such as threshold saturation and that their decoding thresholdsimprove with the repetition factor $q$. We also suggest that the codes asymptotically approach the capacity as $q$ tends toinfinity with any given constituent convolutional code .…

We study time/memory tradeoffs of function inversion: an algorithm, i.e., aninverter, equipped with an s-bit advice on a randomly chosen function $f : n : [n]-n]$ and using $q$ oracle queries to $f$ We make progress on the above intriguing question, both for the adaptive and non-adaptive case, proving the following lower bounds on restricted families of inverters .…

This paper envisions a multi-agent system for detecting the presence of hatespeech in online social media platforms such as Twitter and Facebook . We introduce a novel framework employing deep learning techniques to coordinatethe channels of textual and im-age processing .…

On a standard dataset of Free Sound 50K,comprising of 200 categories, our model outperforms convolutional models . This is significant as unlike in natural language processing and computer vision, we do not perform unsupervised pre-training for outperformingconvolutional architectures . On the same training set, with respect meanaver-age precision benchmarks, we show a significant improvement .…

This paper presents the results of the Russian News Clustering and HeadlineSelection shared task . We propose tasks of Russian newsevent detection, headline selection, and headline generation . The presented datasets for eventdetection and headline selection are the first public Russian datasets for their tasks .…

The proposed method is based on computing multiplespatio-temporal objective quality features on viewports extracted from 360-degree videos . A new model is learnt to properly combine these features into a metric that closely matches subjective quality scores . No individual objective image quality metric always performs the best for all types of visualdistortions, while a learned combination of them is able to adapt to different conditions .…

We learn an interactive vision-based driving policy from pre-recorded driving logs via a model-based approach . A forward model of the world supervises adriving policy that predicts the outcome of any potential driving trajectory . Despite the world-on-rails assumption, the final driving policyacts well in a dynamic and reactive world .…

Tremor is a key diagnostic feature of Parkinson’s Disease (PD), EssentialTremor (ET), and other central nervous system (CNS) disorders . Clinicians or trained raters assess tremor severity with TETRAS scores by observing patients . In this work, we proposed totrain a deep neural network (DNN) with rank-consistent ordinal regression using276 clinical videos from 36 essential tremor patients .…

Gun violence is a severe problem in the world, particularly in the United States . Computer vision methods have been studied to detect guns insurveillance video cameras or smart IP cameras . However, due to no public datasets, it is hard to benchmark how well such methods work in real applications .…

Utilizing clinical texts in survival analysis is difficult because they are largely unstructured . Current automatic extraction models fail to capture textual information comprehensively . They typically require a large amount of data and high-quality annotations for training . In this work, we present a novel method ofusing BERT-based representations of clinical texts as covariates for proportional hazards models to predict patient survival outcomes .…

This work presents the derivation of a model for the heating process of the air of a glass dome, where an indoor swimming pool is located in the bottom of the dome . The problem can be reduced from a three-dimensional to a twodimensional one .…

A conceptual model can be used to manage complexity in both the design and implementation phases of the system development life cycle . This paper discusses model multiplicity (e.g., unifiedmodeling language [UML], model singularity) and a heterogeneous model that involves multiplicity and singularity .…

UniGNN is a unified framework for interpreting the messagepassing process in graph and hypergraph neural networks, which can generalize general GNN models into hypergraphs . Extensive experiments have been conducted to demonstratethe effectiveness of UniGnn on multiple real-world datasets, which outperform the state-of-the-art approaches with a large margin .…

We construct a structure-preserving finite element method and time-steppingscheme for compressible barotropic magnetohydrodynamics . The method preserves thebalance laws governing the evolution of total energy and magnetic helicity, and preserves mass and the constraint $ \operatorname{div}B = 0$ to machine precision .…

Location data are widely used in mobile apps, ranging from location-basedrecommendations, to social media and navigation . But serious privacy concerns arise if users share their locationhistory with the service provider in plaintext . The underlying searchable encryption primitives required to perform the matching on ciphertexts are expensive, and without a proper encoding oflocations and search predicates, the performance can degrade a lot .…

The proposed method is generic enough to be integrated to many variants of weighted ICP without adaptation . It supports localization robustness in harsher terrains, says the authors . The method decreasesthe error by 9.26% in translation and by 21.84% in rotation.…

Rate-Splitting Multiple Access (RSMA) is a flexible and robust multipleaccess scheme for downlink multi-antenna wireless networks . RSMA relies onRate-splitting (RS) at the transmitter and Successive Interference Cancellation (SIC) at receivers . We present the optimal system designsemploying RSMA that target short-packet and low-latency communications as well as robust communications with high-throughput under the practical and importantsetup of imperfect Channel State Information at Transmitter (CSIT) originating from user mobility and feedback latency in the network .…

Massive multiple- input multiple-input multiple-output (MIMO) is one of the key techniques toachieve better spectrum and energy efficiency in 5G system . The channel stateinformation (CSI) needs to be fed back from the user equipment to the basestation in frequency division duplexing (FDD) mode .…

We present two Dialectica-like constructions for models of intensionalMartin-L\”of type theory . We propose a new semantic notion of finite sum for dependent types, generalizingfinitely-complete extensive categories . The second avoids extensivityassumptions using biproducts in a Kleisli category for a fibred additive monad .…

Power control in decentralized wireless networks poses a complex stochasticoptimization problem when formulated as the maximization of the average sumrate for arbitrary interference graphs . Recent work has introduced data-drivendesign methods that leverage graph neural network (GNN) to efficientlyparametrize the power control policy mapping channel state information (CSI) to the power vector .…

A wireless multihop device-to-device (D2D) basedmobile edge computing (MEC) system, where the destination wireless device (WD) is scheduled to compute nomographic functions . We propose a new multi-level over-the-air (OTA) aggregation scheme for the destination WD tocollect the individual partially aggregated intermediate values (IVAs) for reducing communication resource overhead .…

Physical-layer key generation (PKG) can generate symmetric keys between twocommunication ends based on the reciprocal uplink and downlink channels . Bysmartly reconfiguring the radio signal propagation, intelligent reflectingsurface (IRS) is able to improve the secret key rate of PKG . IRS-assistedmultiple-input single-output (MISO) system aims to maximize the secretkey rate by optimally designing the IRS passive beamforming .…

The proposed method can be utilized to train agents in environments with fairly complex state and action spaces . The main challenge of using such a rewardfunction is the high sparsity of positive reward signals . To address this problem, we use a simple prediction-based exploration strategy (called CuriousExploration) along with a Return-based Memory Restoration (RMR) technique which tends to remember more valuable memories .…

In task allocation for real-time domains, alimited number of agents is deployed across a large area to carry out numeroustasks . Agents need to cooperate byforming, disbanding and reforming coalitions . We propose a binary integer program and an anytime and scalableheuristic to solve it .…

A real-time system, we specify the temporal properties on QCTMC by signal temporal logic (STL) To effectivelycheck the atomic propositions in STL, we develop a state-of-art real rootisolation algorithm under Schanuel’s conjecture . Further, we check the generalSTL formula by interval operations with a bottom-up fashion, whose querycomplexity turns out to be linear in the size of the input formula by calling the real root isolation algorithm .…

Lectures were developed for the course Computational SocialChoice of the Artificial Intelligence MSc programme at the University of Groningen . They cover mathematical and algorithmic aspects of voting theory .…

Sequential Recommendation characterizes the evolving patterns by modelingitem sequences chronologically . We use thistransformer to generate fabricated historical items at the beginning of short sequences . We fine-tune the transformer using these augmentedsequences from the time order to predict the next item .…

Bus bunching has become a common phenomenon that undermines the reliability and efficiency of busservices . Little research has focused on bus fleet control due to the tricky characteristic — control action only happens when a busarrives at a bus stop and thus agents do not act simultaneously .…

Distribution matchers for finite alphabets are shown to have informationaldivergences that grow logarithmically with the block length . This is a generalizing abasic result for binary strings .…