Transfer Learning for Human & AI

A thorough backward stability analysis of Hotelling’s deflation, an explicitexternal deflation procedure through low-rank updates for computing manyeigenpairs of a symmetric matrix, is presented . Computable upper bounds of theloss of the orthogonality and the symmetricbackward error norm of the computed eigenvectors are derived .…

Recent advances in deep clustering and unsupervisedrepresentation learning are based on the idea that different views of an inputimage must be closer in therepresentation space . Consensus Clustering usingUnsupervised Representation Learning (ConCURL) improves the clusteringperformance over state-of-the art methods on four out of five image datasets .…

The main goal of NeuralLog is to bridge logic programming and deep learning . The main advantages of neural networks are: to allow neural networks to be defined as logic programs; and to be able to handlenumeric attributes and functions .…

Unlike conventional videos, 360{\deg videos give freedom to users to turntheir heads, watch and interact with content owing to its immersivespherical environment . Although these movements are arbitrary, similarities can be observed between viewport patterns of different users and different videos .…

Quantum computing technologies pose a significant threat to the currently employed public-key cryptography protocols . We discuss the impact of the quantum threat on public key infrastructures (PKIs) We attempt to provide a set of securityrecommendations regarding the PKI from the viewpoints of attacks with quantumcomputers .…

The distributed dual ascent is an established algorithm to solve stronglyconvex multi-agent optimization problems with separable cost functions . In this paper, we study its asynchronouscounterpart . We assume that each agent only relies on the outdated information received from some neighbors .…

Machine Learning model is based on a neural network with several layers of LSTM andfully connected layers . It is trained with historical stock values, technical indicators and Twitter attribute information retrieved, extracted and calculated from posts on the social media platform Twitter .…

Graph Neural Network (GNN) is a promising approach for analyzinggraph-structured data . It has achieved state-of-the-art performances in many tasks, such as node classification, graph matching, clustering, and graphgeneration . As GNNs operate on non-Euclidean data, their irregular data accesspatterns cause considerable computational costs and overhead on conventionalarchitectures such as CPU and CPU .…

Deep neural networks (DNNs) have been ubiquitously applied in many applications . To achieve highmodel coverage with high performance, each accelerator vendor has to develop afull compiler stack to ingest, optimize, and execute the DNNs . To address these issues, this paper proposes an open source framework that enables users to only concentrate on the development of their own code generation tools by reusing as many as possible components inthe existing deep learning compilers .…

We propose Neural Weighted A* a differentiable planner able to produce improved representations of planar maps asgraph costs and heuristics . Training occurs end-to-end on raw images and direct supervision on planning examples . We outperform similar architectures in planning accuracy and efficiency, and can trade offplanning accuracy for efficiency at run-time, using a single, real-valuedparameter .…

We show that for a large class of graphs with suitabledegree and expansion properties with independent exponential mean one edgelengths, there is w.h.p.~a 1-spanner that uses $\approx \frac12n\log n$ edgesand that this is best possible . In particular, our result applies to the randomgraphs $G_{n,p}$ for $np\gg \log n$, in particular .…

There are many deep-learning-based methods with impressive performance for underwater image enhancement, but their memory and model parameter costs are hindrances in practical application . Toaddress this issue, we propose a lightweight adaptive feature fusion network(LAFFNet) The model is the encoder-decoder model with multiple adaptivefeature fusion (AAF) modules .…

The use of video surveillance in public spaces has attracted considerable attention in recent years . But it has been difficult to systematically measure the prevalence and placement of cameras . Here we present a novel approach forestimating the spatial distribution of surveillance cameras .…

The proposed approach uses a map segmentation technique to decompose the environment map into smaller, tractable maps . A simple information gain function is computed to determine the best target region to search during eachiteration of the process . DDQN and A2C algorithms are extended with a stack ofLSTM layers and trained to generate optimal policies for the exploration andexploitation, respectively .…

We propose a weakly-supervised multi-view learning approach to learn category-specific surface mapping without dense annotations . We learn theunderlying surface geometry of common categories, such as human faces, cars,and airplanes, given instances from those categories . Our approach leverages information from multiple views of the object to establish additional consistency cycles, thus improving surface mapping understanding .…

PathBench is a platform for developing, visualizing, training,testing, and benchmarking of existing and future, classical and learned 2D and 3D path planning algorithms . It offers support for Robot Oper-ating System(ROS) and integrates new algorithms with easy and clearly specified .…

Hand Gesture Recognition (HGR) based on inertial data has grown considerably in recent years . The state-of-the-art approaches utilize a single sensor and a vocabulary comprised of simple gestures . In this work we explore the benefits of using multiple inertial sensors .…

This work explores the use of computer vision for image segmentation and classification of medical fluid samples in transparent containers . Handling fluids such as infusion fluids,blood, and urine samples is a significant part of the work carried out in medical labs and hospitals .…

This paper uses a variant of the notion of \emph{inaccessible entropy}(Haitner, Reingold, Vadhan and Wee, STOC 2009) to give an alternative construction and proof for the fundamental result, first proved by Rompel (STOC1990) that UOWHFs can be based on any one-way functions .…

Most video super-resolution methods focus on restoring high-resolution videoframes from low-resolution videos without taking into account compression . Most videos on the web or mobile devices are compressed, and the compression can be severe when the bandwidth is limited . In this paper, we propose a new compression-informed video-super-resolution model to restorehigh-resolution content without introducing artifacts caused by compression .…

This paper considers the data association problem for multi-target tracking . Multiple hypothesis tracking is a popular algorithm for solving this problem . It is NP-hard and is is quite complicated for tracking a large number of targets or tracking maneuvering targets .…

Many users, especially older adults, are not aware of accessibility features or do not know which combination to use . Automated recommendation could help people find beneficial accessibility features, authors say . Their work demonstrates the need to increase awareness of existing accessibility features on mobile devices .…

In this work, we introduce a convex optimization strategy to train quantized NNs with polynomial activations . Our method leverages hiddenconvexity in two-layer neural networks from the recent literature, semidefinitelifting, and Grothendieck’s identity . We show that certainquantized NN problems can be solved to global optimality in polynnomine-time in Polynomial-time .…

Machine Learning (ML) algorithms, especially Deep NeuralNetwork (DNN) models, become a widely accepted standard in many domains moreparticularly IoT-based systems . DNN models reach impressive performances in several sensitive fields such as medical diagnosis, smart transport or securitythreat detection, and represent a valuable piece of Intellectual Property .…

Pre-trained text encoders have drawn sustaining attention in natural languageprocessing (NLP) They have shown their capability in obtaining promising results indifferent tasks . We propose topre-train n-gram-enhanced Encoders with a large volume of data and advanced techniques for training . We try to extend the encoder to different languages as well as different domains, where it is confirmed that the samarchitecture is applicable to these varying circumstances and new state-of-the-art performance is observed from a long list of NLP tasks across the languages and domains .…

Authors transpose a discrete notion of indetermination coupling in thecase of continuous probabilities . They show that this coupling, expressed ondensities, cannot be captured by a specific copula which acts on cumulativedistribution functions without a high dependence on the margins .…

We present a streaming end-to-end speech recognition model based on Monotonic Chunkwise Attention (MoCha) jointly trained with enhancement layers . We introduce two training strategies: GradualApplication of Enhanced Features (GAEF) and Gradual Reduction of Enhanced Loss (GREL) With GAEF, the model is initially trained using clean features .…

Industrial Internet-of-Things (IIoT) facilitates efficiency and robustness in factory operations . The synergy of complex event processing (CEP) and machinelearning (ML) has been developed actively in the last years in IIoT to identify patterns in heterogeneous data streams and fuse raw data into tangible facts .…

We present a more general analysis of $H$-calibration for adversariallyrobust classification . By adopting a finer definition of calibration, we cancover settings beyond the restricted hypothesis sets studied in previous work . We both fix some previous calibration results (Bao et al.,…

The model improvescoherence by exploiting neural word embeddings through a graph-based clusteringmethod . Unlike typical topic models, this approach works without knowing the true number of topics . Experimental results on the real-life multimodal dataset MuSe-CaR demonstrates that our approach extracts coherent and meaningfultopics, outperforming baseline methods .…

The Eigenfactor is a journal metric, which was developed by Bergstrom and his colleagues at the University of Washington . It establishes the importance, influence or impact of a journal based on its location in a network of journals . While journal-basedmetrics have been criticized, it has also been suggested as analternative in the widely used San Francisco Declaration on Research Assessment(DORA) The algorithm is based on Eigenvector centrality, i.e.…

C. elegans shows chemotaxis using klinokinesis where the worm senses the concentration based on a single concentration sensor to compute the concentration gradient to perform foraging through gradient ascent/descenttowards the target concentration . The biomimeticimplementation requires complex neurons with multiple ion channel dynamics aswell as interneurons for control .…

We consider the strong secret key (SK) agreement problem for the satellitecommunication setting . Legitimate receivers have access to an authenticated, noiseless, two-way, and public communication link . The noise variances for Alice’s and Bob’s measurement channels are both fixed to a value $Q1$, whereas the noise over Eve’s measurement channel has a unitvariance, so $Q$ represents a channel quality ratio .…

Citadel is a scalable collaborative ML system that protects the privacy of both data owner and model owner in untrusted infrastructures . Citadel performs distributed training across multiple training enclaves running on behalf of data owners and an aggregatorenclave . Citadel scales to alarge number of enclaves with less than 1.73X slowdown caused by SGX .…

We study the meta-learning of numerical algorithms for scientific computing . It combines the mathematically driven, handcrafted design of general algorithm structure with a data-driven adaptation to specific classes of tasks . This represents a departure from the classical approaches in numericalanalysis, which typically do not feature such learning-based adaptations .…

This study analyzed the performance of different machine learning methods for winter wheat yield prediction using extensive datasets of weather, soil, and phenology . The results indicated that nonlinear models such as deep neuralnetworks (DNN) and XGboost are more effective in finding the functionalrelationship between the crop yield and input data compared to linear models .…

Machine reading comprehension is a sub-field in natural languageprocessing or computational linguistics . MRC aims to help computers understandunstructured texts and then answer questions related to them . The best model obtains an F1 score of 45.27%, which is 30.91 points behind human performance (76.18%), indicating that there is ample room for improvement.…

The chromatic index of a cubic graph is either 3 or 4 . Edge-Kempe switching can be used to transform edge-colorings of cubic graphs . It is further connected to cycle switching of Steiner triple systems, for example, for which an improvement of the classification algorithm is presented .…

An infinite word is an infinite Lyndon word if it is smaller, with respect to lexicographic order, than all its proper suffixes . An endomorphisms generating Lyndon infinite words are provided . A characterization ofbinary endorphisms generating infinite words is provided .…

Commonsense-inference Augmented neuralStoryTelling (CAST) is a framework for introducing commonsense reasoning into the generation process while modeling the interaction between multiple characters . We find that our CAST method produces significantly more coherent and on-topictwo-character stories, outperforming baselines in dimensions including plotplausibility and staying on topic .…

The anthology of spoken languages today is inundated with textual information, necessitating the development of automatic summarization models . In this manuscript, we propose an extractor-paraphraser based abstractivesummarization system that exploits semantic overlap as opposed to itspredecessors that focus more on syntactic information overlap .…

Hadamard matrices are square $n\times n$ matrices whose entries are ones and minus ones . Matrices are orthogonal to each other with respect to the standard scalar product in $\Bbb R^n$. Each Hadamar matrix can be transformed to a matrix whose entries were zeros and ones .…

There are numerous e-assessment systems devoted to specific domains underdiverse license models . Cost, extensibility, and maintainability are relevant issues for an institution . Ease of use and inclusion into courses are main concerns . For students the user experience and fast transparent feedback plus “better” tests are most important .…

We consider the infinite-horizon, discrete-time full-information controlproblem . Motivated by learning theory, as a criterion for controller design we focus on regret . In thefull-information setting, there is a unique optimal non-causal controller that dominates all other controllers . The regret-optimal controller is the sum of the classical $H_2$ state-feedback law and a finite-dimensional controller obtainedfrom the Nehari problem .…

We consider non-cooperative facility location games where both facilities and clients act strategically and heavily influence each other . This contrasts established game-theoretic facility location models with non-strategic clientsthat simply select the closest opened facility . We focus on a natural client behavior similar to classical loadbalancing: our selfish clients aim for a distribution that minimizes their maximum waiting times for getting serviced, where a facility’s waiting timecorresponds to its total attracted client weight .…

A unified system integrating a compact object detector and a surroundingenvironmental condition classifier for enhancing the robustness of objectdetection scheme in advanced driver assistance systems (ADAS) is proposed in this paper . The proposed system includes two main components: (1) a compactone-stage object detector which is expected to be able to perform at acomparable accuracy compared to state-of-the-art object detectors, and (2) an environmental condition detector that helps to send a warning signal to the cloud in case the self-driving car needs human actions due to the significance of the situation .…

This paper analyzes the generalization error of two-layer neural networks for computing the ground state of the Schr\”odinger operator on a $d$-dimensionalhypercube . We prove that the convergence rate of the error isindependent of the dimension $d$, under the a priori assumption that the groundstate lies in a spectral Barron space .…

Deep learning-based optical flow techniques do not perform well for non-rigidmovements such as those found in faces . We hypothesize that learning opticalflow on face motion data will improve the quality of predicted flow on faces . The performance of FlowNetS trained on face images surpassed that of other opticalflow CNN architectures, demonstrating its usefulness.…

An intelligent zero trust architecture (i-ZTA) as a security framework in 5G/6G networks with untrusted components . We introduce key ZT principles as real-time Monitoring of thesecurity state of network assets, Evaluating the risk of individual access requests, and Deciding on access authorization using a dynamic trust algorithm,called MED components .…

The main challenge faced by policymakers is to learn a policy from scratch by interacting with a complex environment in a few trials . This work introduces apractical, data-efficient policy learning method, named Variance-Bonus MonteCarlo Tree Search~(VB-MCTS) It can copy with very little data andfacilitate learning from scratch in only a few trial times .…