Transfer Learning for Human & AI

MDETR is an end-to-end modulated detector thatdetects objects in an image conditioned on a raw text query . We use a transformer-based architecture to reason jointly over text and image by fusing the two modalities at an early stage of the model .…

Large collections of observatory publications seem to be rare; or at the least rarely digitally described or accessible on the Internet . Notable examples to the contrary are the WoodmanAstronomical Library at Wisconsin-Madison and the Dudley Observatory in Loudonville, New York both in the US .…

Android is a mobile operating system based on a modified version of the Linux kernel and other open source software . It is an operating system forlow powered devices that run on battery and are full of hardware like GlobalPositioning System (GPS) receivers, cameras, light and orientation sensors, Wi-Fi and LTE (4G telephony) connectivity and a touch screen .…

A low-complexity convolutional neural network estimator which learns theminimum mean squared error channel estimator for single-antenna users was recently proposed . We show that by using discrete Fourier transform based pilots the number oflearnable network parameters decreases significantly and the online run time of the estimator is reduced considerably .…

Radio frequency (RF) wireless energy transfer (WET) is a key technology that may allow seamlessly powering future low-energy Internet of Things(IoT) networks . To enable efficient massive WET, channel state information(CSI)-limited/free multi-antenna transmit schemes have been recently proposedin the literature .…

Many approximate GMR algorithms have been proposed in the past decades, although none of them provides optimality guarantees . We discuss the importance of the choice of the dissimilarity measure and the issue of consistency of all steps of a reduction algorithm with the chosen measure .…

Android is a mobile operating system based on a modified version of the Linux kernel and other open source software . It is an operating system forlow powered devices that run on battery and are full of hardware like GlobalPositioning System (GPS) receivers, cameras, light and orientation sensors, Wi-Fi and LTE (4G telephony) connectivity and a touch screen .…

Human learning benefits from multi-modal inputs that often appear as richsemantics . This enables us to learn generalizable concepts from very limited visual examples . However, current few-shot learning (FSL) methods use numerical class labels to denote object classes which do not provide rich semantic meanings about the learned concepts .…

The Alternating DirectionMethod of Multipliers (ADMM) through the concept of consensus variables is apractical algorithm in this context . In this paper, westudy the effect of the local data sets of users in the distributed learning ofADMM. Our aim is to deploy variational inequality (VI) to attain an unifiedview of ADMM variations.…

Adversarial examples mainly exploit changes to input pixels to which humans are not sensitive to . We show that it is tightly linked to the spatial frequencycharacteristics of the data at stake . Depending on the data set, the same constraint may results in very different level of robustness (up to 0.41adversarial accuracy difference) To explain this phenomenon, we conduct several experiments to enlighten influential factors such as the level ofsensitivity to high frequencies, and the transferability of adversarialperturbations between original and low-pass filtered inputs .…

Database systems analyze queries to determine upfront which data is needed for answering them and use indexes and other physical design techniques to speed-up access to that data . For important classes of queries, e.g.,HAVING and top-k queries, it is impossible to determine up-front what data is relevant .…

Growing interest in open source software has also been attributed to developers deciding to use their technical skills to benefit a commonsocietal good . Researchers conducted 21semi-structured interviews with OSS for Social Good (OSS4SG) contributors . They found that OSS4SG contributors focus less on benefiting themselves by padding their resumewith new technology skills and are more interested in leaving their mark on society at statistically significant levels .…

A sphere is a fundamental geometric object widely used in (computer aided)geometric design . It possesses rational parameterizations but no parametricpolynomial parameterization exists . The present study provides an approach to the optimal approximation of equilateral spherical triangles by parametric polynomial patches if the measure of quality is the (simplified) radial error .…

Deformations of viscoelastic materials such as soft tissues, metals at hightemperature, and polymers can be described as Volterra integral equations of the second kind . We use a spatially discontinuous Galerkin finite element method and a finitedifference method in time to formulate the fully discrete problem .…

In this paper, we propose a fresh perspective on argumentation semantics, toview them as a relational database . It offers encapsulation of the underlying argumentation graph, leading to the concept of relational semantics . We show that many existingsemantics such as explanation semantics, multi-agent semantics, and moretypical semantics, that have been proposed for specific purposes, are understood in the relational perspective .…

A falsification algorithm for neuralnetworks directs the search for a counterexample guided by a safetyproperty specification . We evaluate our algorithm on 45 trained neural networkbenchmarks of the ACAS Xu system against 10 safety properties . We show that our procedure detects all the unsafe instances that other tools also report as unsafe .…

ECLIPSE is a spatio-temporal neural network architecture that models cloud motion from sky images to predict both future segmented images and corresponding irradiance levels . It is based on the analysis of sequences of ground-taken sky images . It reduces temporal delay while generating visually realistic futures .…

VGG models used two sets of fully connected layers for the classification part of their architectures, which significantly increasesthe number of models’ weights . ResNet and next deep convolutional models used the Global Average Pooling (GAP) layer to compress the feature map and feed it to the classification layer .…

ODDObjects is designed to detect anomalies of various categories using unsupervised autoencoders trained on COCO-style datasets . Themethod utilizes autoencoder-based image reconstruction, where highreconstruction error indicates the possibility of an anomaly . The frameworkextends previous work on anomaly detection with autoenoders, comparing state-of-the-art models .…

Traditional video summarization methods generate fixed video representations regardless of user interest . Traditional methods limit users’ expectations in content search and exploration scenarios . In this work, a new method is proposed that uses a specialized attention network andcontextualized word representations to tackle this task .…

In ML-aided decision-making tasks, the human-in-the-loop prefers high-level concept-based explanations instead of low-levelexplanations based on model features . We leverage multi-task learning to train a neural network that learns to predict a decision task based on the predictions of aprecedent explainability task .…

Adaptive Bit Rate (ABR) decision plays a crucial role for ensuringsatisfactory Quality of Experience (QoE) in video streaming applications . Past network statistics are mainly leveraged for future network bandwidthprediction . This paper proposes to learn the ANT (a.k.a., Accurate Network Throughput) model to characterize the full spectrum of network throughput dynamics in the past forderiving the proper network condition associated with a specific cluster ofnetwork throughput segments (NTS) Each cluster of NTS is then used to generate a dedicated ABR model, by which we wish to better capture the network dynamicsfor diverse connections .…

The promise constraint satisfaction problem (PCSP) is a recently introduced generalisation of the constraint satisfaction (CSP) that capturesapproximability of satisfiable instances . A PCSP instance comes with two formsof each constraint: a strict one and a weak one . Given the promise that asolution exists using the strict constraints, the task is to find a solutionusing the weak constraints .…

Automatic algorithm-hardware co-design for DNN has shown great success in improving the performance of DNNs on FPGAs . The solution searched by our algorithm achieves 72.5% top-1 accuracy on ImageNet at framerate 50, which is 60% faster than MnasNet . With lowcomputational cost, our algorithm can generate quantized networks that achievestate-of-the-art accuracy and hardware performance on Xilinx Zynq (ZU3EG) FPGA for image classification on image classification .…

The method significantly outperforms a state-of-the-art approach for scene compression, achieving simultaneously higherquality reconstructions and lower bitrates . The function is implemented as a neural network and jointly trained for reconstruction as well as compression, in an end-to-end manner, with the use of an entropy penalty on the parameters .…

Gastric diffuse-type adenocarcinoma represents a disproportionately highpercentage of cases of gastric cancers occurring in the young . Usually it affects the body of the stomach,and presents shorter duration and worse prognosis compared with thedifferentiated (intestinal) type of adenodecarinoma . The main difficultyencountered in the differential diagnosis occurs with the diffuse type.…

A synthetic data generation pipeline that generates arbitrary amount of 3D data and annotated building annotations . The pipeline is designed to be modular, modular and adaptable to a large variety of tasks . All code and data are made publicly available to researchers and the pipeline is free to download and use for training purposes in the wild .…

We present an application of invariant polynomials in machine learning . We find a reduction of the loss on training data and a significant reduction on validation data . We discuss and prove some theorems to make use of these invariant generators in machinelearning algorithms in general and in neural networks specifically .…

Mutual Contrastive Learning (MCL) is a collaborative learning method for general visual representation learning . The core idea of MCL is to perform mutual interaction and transfer of contrastive distributions among acohort of models . Benefiting from MCL, each model can learn extra contrastiveknowledge from others, leading to more meaningful feature representations for recognition tasks .…

Continual zero-shot learning(CZSL) is a new domain to classify objects sequentially the model has not seen during training . It is more suitable thanzero-shot and continual learning approaches in real-case scenarios when datamay come continually with only attributes for a few classes and attributes andfeatures for other classes .…

Binaural sound that matches the visual counterpart is crucial to bring immersive experiences to people in augmented reality (AR) and virtual reality (VR) applications . This paper proposes Points2Sound, amulti-modal deep learning model which generates a binaural version from monoaudio using 3D point cloud scenes .…

Session types are becoming popular and have been integrated in several programming languages . The notion of subtyping used in session type implementations is the one defined by Gay and Hole for synchronous communication . The aim of this paper is to make the growing body of knowledge about asynchronous session subtypp more accessible to non-experts and promote its integration in practical applications of sessiontypes .…

The dissertation proposes the use of a multi-objective optimization framework for designing and selecting among enhanced GOP configurations in videocompression standards . The proposed methods achieve fine optimization over aset of general modes that include: (i) maximum video quality, (ii) minimumbitrate, (iii) maximum encoding rate (previously minimum encoding time mode) and (iv) can be shown to improve upon the YouTube/Netflix default encoder modesettings over a set of opposing constraints to guarantee satisfactoryperformance .…

The steady-state visual evoked potential (SSVEP) is one of the most widelyused modalities in brain-computer interfaces (BCIs) The existence of harmonics and the limited range of responsiverequencies in SSVEP make it challenging to further expand the number of targets presented .…

Timed Automata (TA) are a very popular modeling formalism for systems with time-sensitive properties . A common task is to verify if a network of TAsatisfies a given property, usually expressed in Linear Temporal Logic (LTL) The produced CLTLoc formula can then be solved by toolssuch as Zot, which transforms CLTLOC properties into the input logics of SMT solvers .…

Hybrid additive manufacturing techniques likethermoforming are becoming popular for prototyping freeform electronics . 3Dprinting the sheet material allows embedding conductive traces within thinlayers of the substrate, which can be vacuum-formed but remain conductive andinsulated . We characterise the behaviour of the vacuum-forming 3D printed sheet .…

There is a limited evidence-based and thorough understanding of the types of issues faced by microservices system developers and causes that trigger the issues . Technicaldebt (321), Build (145), Security (137) and Serviceexecution and communication (119) are prominent . “General programming errors”, “Poor security management”, “invalidconfiguration and communication”, and “Legacy versions, compatibility anddependency” are the predominant causes for the leading four issue categories .…

Reliable and frequent population estimation is key for making policies aroundvaccination and planning infrastructure delivery . Since censuses lack thespatio-temporal resolution required for these tasks, census-independentapproaches, using remote sensing and microcensus data, have become popular . We assess the feasibility of using publicly availabledatasets to estimate population .…

Android is a mobile operating system based on a modified version of the Linux kernel and other open source software . It is an operating system forlow powered devices that run on battery and are full of hardware like GlobalPositioning System (GPS) receivers, cameras, light and orientation sensors, Wi-Fi and LTE (4G telephony) connectivity and a touch screen .…

An extended collection and processing of information regarding charging of electric vehicles may turn each electric vehicle charging station into a valuable source of streaming data . Chargingpoint operators may profit from all these data for optimizing their operationand planning activities .…

The study compares the performance of five types of measures: age, sex, social, health-related, biomarker and genetic single nucleotidepolymorphisms (SNPs) The predicted outcome variable was limiting long-termillness one and five years from baseline . Health-related measures had the strongest prediction of future health status, with genetic data performing poorly .…

Deep models have shown their vulnerability when processing adversarialsamples . As for the black-box attack, training a substitute model for adversarialattacks has attracted wide attention . Previous substitute training approaches focus on stealing the knowledge of the target model based on real training data or synthetic data .…

This paper is inspired by the PQ penny flip game . It employs group-theoretic concepts to study the original game and also its possible extensions . We show that the game can be played in the all dihedral groups $D_{8 n}$, $n \geq 1$ with any significant change .…

We provide a counterexample to a conjecture by B. Connelly about density of circle packings .…

We present a framework that allows for the non-asymptotic study of the $2$-Wasserstein distance between the invariant distribution of an ergodicstochastic differential equation and the distribution of its numericalapproximation in the strongly log-concave case . This allows us to study in aunified way a number of different integrators proposed in the literature for the overdamped and underdamped Langevin dynamics .…

The performance of applications rely on high-quality knowledge bases, a.k.a. Knowledge Graphs . To ensure their quality one important task is KnowledgeValidation, which measures the degree to which statements or triples of a Knowledge Graph (KG) are correct . We propose and implement a validation approach that computes a confidence score for everytriple and instance in a KG .…

This paper dwells on certain novel game-theoretic investigations inbio-mimicry . The model is used to study the situation where multi-armedbandit predators with zero prior information are introduced into the ecosystem . The prey can be either nutritious or toxic to the predator, but the prey may signal (possibly) deceptively without revealing its true “type” The model uses a model to study a panmictic ecosystem occupied by species of prey with a relatively short lifespan, which evolve mimicry signals over generations .…

This work studies finite sample approximations of the exact and entropic regularized Wasserstein distances between centered Gaussian processes and, moregenerally, covariance operators of functional random processes . We first show that these distances/divergences are fully represented by reproducing kernelHilbert space (RKHS) covariance and cross-covariance operators associated with the corresponding covariance functions .…

Bayesian optimization (BO) with Gaussian processes is a powerful methodology to optimize an expensive black-box function . The expected improvement (EI) and probability of improvement (PI) are among the most widely used schemes for BO . The proposed method isnumerically inexpensive, is easy to implement, can be easily parallelized, and on benchmark tasks shows a performance superior to EI and GP-UCB .…