Transfer Learning for Human & AI

Photonics Crystal (PhC) nanocavities are $mu m^2 scale devices offering 100fJ switching operation under picoseconds-scale switching speed . Stochastic computing allows a drastic reduction in hardware complexity usingserial processing of bit streams . The proposed architecture leads to 8.5nJ/pixel energy consumption and 512ns/pixel processing time .…

Edge nodes are crucial for detection against multitudes of cyber attacks on Internet-of-Things endpoints . The resource constraints in this novel network infrastructure tier constricts the deployment of existing Network Intrusion Detection System withDeep Learning models (DLM) We address this issue by developing a novel light,fast and accurate ‘Edge-Detect’ model, which detects Distributed Denial ofService attack on edge nodes using DLM techniques .…

Perceptions of system competence and communicative ability, termed partnermodels, play a significant role in speech interface interaction . Yet we do not know what the core dimensions of this concept are . Taking a psycholexicalapproach, our paper is the first to identify the key dimensions that definepartner models in speech agent interaction .…

AttentionFlow is a new system to visualise networks of time series and the dynamic influence they have on one another . The system simultaneously presents the time series on each node using two visual encodings: a tree ring for anoverview and a line chart for details .…

In a collaborative-filtering recommendation scenario, biases in the data will likely propagate in the learned recommendations . We propose NAECF, a conceptually simple but effective idea to address this bias . The idea consists of adding an autoencoder (AE)layer when learning user and item representations with text-based ConvolutionalNeural Networks .…

The intertwined processes of learning and evolution in complex environmentalniches have resulted in a remarkable diversity of morphological forms . Many aspects of animal intelligence are deeply embodied in theseevolved morphologies . The principles governing relations between environmental complexity, evolved morphology, and the learnability of intelligent control remain elusive, partially due to the substantial challenge of performing large-scale in silico experiments on evolution and learning .…

We consider optimal intervention in the Elliott-Golub-Jackson network model . We show that it can be transformed into an influence maximization problem,interpreted as the reverse of a default cascade . We prove several results about optimal intervention: it is NP-hard and additionally hard toroximate to a constant factor in polynomial time .…

New York has become one of the worst-affected COVID-19 hotspots and apandemic epicenter due to ongoing crisis . The city saw its first case on March 1, 2020, but disruptions in mobility can be seen only after shelter in place orders was put in effect .…

We formulate the search for phenomenological models of synaptic plasticity as an optimization problem . We employ Cartesian genetic programming to evolvebiologically plausible human-interpretable plasticity rules that allow a given network to successfully solve tasks from specific task families . We demonstrate that the evolved rules perform competitively with known solutions .…

In Speech Emotion Recognition (SER) emotional characteristics often appear in diverse forms of energy patterns in spectrograms . Typical attention neuralnetwork classifiers of SER are usually optimized on a fixed attentiongranularity . In this paper, we apply multiscale area attention in a deepconvolutional neural network to attend emotional characteristics with variedgranularities .…

Confusion2vec, motivated from human speech production and perception, is a word vector representation whichencodes ambiguities present in human spoken language in addition to semanticsand syntactic information . Word vector representations enable machines to encode human language forspoken language understanding and processing .…

We study the problem of learning Ising models satisfying Dobrushin’s condition in the outlier-robust setting where a constant fraction of the samples are adversarially corrupted . Our main result is to provide the firstcomputationally efficient robust learning algorithm for this problem withnear-optimal error guarantees .…

In many methods of sound event detection (SED) a segmented time frame is considered as one data sample to model training . The durations of sound events depend on the sound event class, e.g. The difference in the duration between sound event classes results in a serious data imbalance in SED .…

Session-based recommendation aims to predict user’s next behavior from current session and previous anonymous sessions . Self-attention networks (SR-SAN) capture the global dependencies among all items of a session . In SR-SAN, a single item latent vector is used to capture both current interest and global interest instead of session embedding .…

Modelling tools perform instantaneous cycle detection(ICD) on composite models having feedback loops to reject the models if the feedback loops are mathematically unsound . The cycle detection problem becomes harder when the model’s input to output dependencies are mode-dependent, i.e. changes for certain events generated internally or externally as inputs .…

QuizCram shows users a question to answer, with an associated videosegment . Users can use these questions to navigate through video segments, and find video segments they need to review . Users practice answering and reviewing questions more when using Quizcram, and are better able to remember answers to questions they encountered .…

In multi-agent reinforcement learning, the problem of learning to act is difficult because the policies of co-players may be heavilyconditioned on information only observed by them . On the other hand, humans form beliefs about the knowledge possessed by their peers and leveragebeliefs to inform decision-making .…

We have built a browser extension that teaches vocabulary to users in the context of Facebook feeds and arbitrary websites . On Facebook, quizzes show up as part of the news feed, while on other sites, the quizzes appear where advertisements normally would .…

In this paper, we revisit the classic CountSketch method, which is a sparse,random projection that transforms a (high-dimensional) Euclidean vector $v$ to a vector of dimension $2t-1) s . We also study the variance in the setting where an innerproduct is to be estimated from two CountSkeches .…

The proposed method allows us to adjust three target-colors chosen by a user in an input image so that each target color is the same as the corresponding destination (benchmark) one . Incontrast, white balancing is a typical technique for reducing the colordistortions, however, they cannot remove lighting effects on colors other than white .…

In remote regions (e.g., mountain and desert), cellular networks are usuallysparsely deployed or unavailable . Offloading tasks to urban terrestrial cloud via satellite link will lead to high delay . The proposed Deep Reinforcement learning-based Task Offloading (DRTO) algorithm can accelerate learning process by adjusting the number of candidate locations .…

Distributed caching systems such as content distribution networks oftenadvertise their content via lightweight approximate indicators (e.g., Bloomfilters) We focus onfalse-negatives induced by indicator staleness, which arises whenever the system advertises the indicator only periodically, rather than immediately reporting every change in the cache .…

Quantum Technologies is a term that is getting broader with every passingyear . Nanotechnology and electronics operate in this realm . Here we taxonomizeand analyze 48,577 patents in this area from 2015 to present captured with acomprehensive query in Relecura’s patent database .…

Rule-based models, such as decision trees, are conventionally deemed to be the most interpretable AI . Recent work uses propositional satisfiability (SAT) solving (and itsoptimization variants) to generate minimum-size decision sets . The approach makes use of modern maximum satisfiability and integer linear programming technologies .…

The energy consumptionpattern across the appliances, houses, communities and entire cities help energy utility companies and consumers plan their electricity generation and consumption . The edge or connection represents energy flow between two participants of the network, these connections last till the power is being consumed/generated .…

Current methods that combine both audio and visual information use 2D representations such as images to guide theseparation process . This paper proposes a multi-modal deep learning model to performmusic source separation conditioned on 3D point clouds of music performancerecordings .…

The absorption and re-radiation energy from molecules in the air can influence the Multiple Input Multiple Output (MIMO) performance in high-frequency bands, e.g., millimeter wave (mmWave) andterahertz . Some common atmosphere molecules, such as oxygen andwater, can absorb and .re-radiate energy in their natural resonance frequencies, such .…

Pecan is capable of efficiently proving non-trivial mathematical theorems about allSturmian words . It is an important object in combinatorics on words .…

This paper presents a self-supervised learning framework, named MGF, forgeneral-purpose speech representation learning . We propose to usegenerative learning approaches to capture fine-grained information at smalltime scales . For phoneme-scalelearning, we borrow idea from the masked language model but tailor it for thecontinuous speech signal by replacing classification loss with a contrastiveloss .…

The most expensive task of our evolutionary algorithm is the evaluation of candidatepaths . We implemented gradient boostingdecision trees to classify candidate paths in order to identify good candidates . The cost of the not-so-good candidates is simply forecasted. The computational performance was significantly improved at the expense of a limited loss of accuracy .…

Sentiment Analysis (SA) software systems may exhibit unintended demographic bias based on specific characteristics (e.g., gender, occupation, country-of-origin, etc.) Such biases manifest in an SAsystem when it predicts a different sentiment for similar texts that differ only in the characteristic of individuals described .…

This chapter describes how to use intersection and closest-hit shaders to implement real-time visualizations of complex fractals using distancefunctions . The Mandelbulb and Julia Sets are used as examples .…

We develop a new framework for analyzing importance sampling methods in regression problems . We show that the query complexity ofleast absolute deviation regression is $Theta(d/\epsilon^2)$ up to logarithmicfactors . We introduce the notion of robust uniform convergence, which is a new guarantee for the empirical loss .…

Using virtual reality setups, users can fade out of their surroundings and dive fully into a thrilling and appealing virtual environment . Gait-based interactions, using the variety of information contained in human gait, could offer interesting benefits for immersive experiences .…

LoRa wireless technology is an increasingly prominent solution for massiveconnectivity and the Internet of Things . Stochastic geometry and numerical analysis of LoRa networks usually consider uniform end-device deployments . Realdeployments however will often be non-uniform, for example due to mobility .…

LoRa networks have been deployed all over the world and are a major enabling technology for the Internet of Things . Massive connectivity applications such as smart metering, agriculture, and supply chain \& logistics are most suitable for LoRa deployments due to their long range, low cost, and low power features .…

Weakly labelled audio tagging aims to predict the classes of sound events within an audio clip, where the onset and offset times of the sound events are not provided . Previous works have used the multiple instance learning (MIL)framework, and exploited the information of the whole audio clip by MIL pooling functions .…

Recent research shows that applying individualinteractions and movements data could help managing the pandemic thoughmodelling the spread of infectious diseases on social contact networks . Infectious diseases are a significant threat to human society which was oversighted before the incidence of COVID-19, although according to the report of the World Health Organisation (WHO) about 4.2 million people die annually dueto infectious disease .…

Internet of Things (IoT) applications drive the behavior of IoT deployments according to installed sensors and actuators . IoTSeer uncovers undesired states caused by physicalinteractions caused by design flaws or malicious intent . We use the security tool in an actual house with 13actuators and six sensors with 37 apps and demonstrate its effectiveness andperformance .…

InfoColorizer provides flexibility by considering users’ preferences, and tailors suggested palettes to the spatiallayout of elements . We build a recommendation engine by utilizing deep learningtechniques to characterize good color design practices from data . We conducted a comprehensive four-part evaluation, including casestudies, a controlled user study, a survey study, and an interview study .…

Unmanned Aerial vehicles (UAVs) are widely used as network processors in mobile networks, but more recently, UAVs have been used in Mobile EdgeComputing as mobile servers . The contributions of our work include: optimizing the quality of service for mobile edge computing and path planning in the samereinforcement learning framework .…

An important aspect of AI design and ethics is to create systems that reflect aggregate preferences of the society . We propose a new social choice function inspired by PageRank algorithm . The function ranks voting options based on the Condorcet graph of pairwise comparisons .…

A standard method for speaker verification consists of extracting speaker embeddings with a deep neuralnetwork and processing them through a backend composed of probabilistic lineardiscriminant analysis (PLDA) and global logistic regression score calibration . This method is known to result in systems that work poorly on conditionsdifferent from those used to train the calibration model .…

Theaverage treatment effect (ATE) of campaign strategies need to be monitored throughout the campaign . A/B testing is usually conducted for such needs, but the existence of user interaction can introduce interference to normal testing . With the help of link prediction, LinkLouvain design a way to minimize graph interference and it gives an accurate andsound estimate of the campaign’s ATE .…

Cross-lingual voice conversion (VC) is an important and challenging problem due to significant mismatches of the phonetic set and the speech prosody of different languages . We build upon the neural text-to-speech (TTS) model, i.e., FastSpeech, and LPCNet neural vocoder to design a new cross-language VC framework .…

In this paper, a Deep Q-Network (DQN) based multi-agent multi-user powerallocation algorithm is proposed for hybrid networks composed of radiofrequency (RF) and visible light communication (VLC) access points . The DQN-based algorithm converges to the desired user rate in halfduration on average while converging with the rate of 96.1% compared to the Q-Learning (QL) based algorithm’s convergence rate of 72.3% .…

Deep complex U-Net structure and convolutional recurrent network (CRN) achieve state-of-the-art performance for monaural speech enhancement . The CCBAM is a lightweight and general module which can be easilyintegrated into any complex-valued convolutionsal layers . We further propose a mixed loss function to jointly optimize the complex models in both time-frequency (TF) domain and time domain .…

In this paper, we propose a novel method that trains pass-phrase specificdeep neural network (PP-DNN) based auto-encoders for creating augmented datafor text-dependent speaker verification (TD-SV) The method improves the performance for both conventional feature and deep bottleneck feature using both Gaussian mixture model- universal background model (GMM-UBM) and i-vector framework .…

A novel artificial cognitive mapping system using generative deepneural networks (VAE/GAN) can map input images to latent vectors andgenerate temporal sequences internally . The results show that the distance of the predicted image is reflected in the length of the corresponding latent vector after training .…

Bayesian Neural Networks (BNN) have recently emerged in the Deep Learning world for dealing with uncertainty estimation in classification tasks . They are used in many application domains such as astrophysics, autonomous driving and astrophysics . In this paper, we present a new method called “Dropout Regulation”(DR) which consists of automatically adjusting the dropout rate during training using a controller as used in automation.DR…