Transfer Learning for Human & AI

Many real-world applications require to convert speech files into text with limited resources . This paper proposes a method to recognizelarge speech database fast using the Transformer-based end-to-end model . The proposed system can convert speeches into text in less than 3 minutes with 10.73% character error rate which is 27.1% relatively low compared to conventional DNN-HMM based recognition system .…

In this paper we the formulation of inverse problems as constrainedminimization problems and their iterative solution by gradient or Newton type . We carry out a convergence analysis in the sense of regularization methods and discuss applicability to the problem of identifying the spatially varyingdiffusivity in an elliptic PDE from different sets of observations .…

Deep ensembles perform better than a single network thanks to diversity among their members . New training criterion called DICE increases diversity by reducing spurious correlations among features . The main idea is that features extracted from pairs of members should only share information useful for target class prediction without being conditionally redundant .…

We study the multi-agent safe control problem where agents should avoid collisions to static obstacles and collisions with each other while reaching their goals . We propose a novel joint-learning framework that can be implemented in a decentralizedfashion . Such adecentralized framework can adapt to an arbitrarily large number of agents .…

In this work, the Bayesian approach to inverse problems is formulated in anall-at-once setting . This method naturally results in a whole posteriordistribution for the unknown target, not just point estimates . We analyze the degree ofill-posedness and the convergence of the method .…

We consider the problem of designing codes with flexible rate (referred to asrateless codes), for private distributed matrix-matrix multiplication . A masterserver owns two private matrices and hires workernodes to help computing their multiplication . The size of the tasks, hence the rate of the scheme, depends on the number of workers that themaster waits for .…

Performance-influence models can help stakeholders understand how and whereconfiguration options and their interactions influence the performance of asystems . Current black-box techniques combine various sampling and learning strategies, resulting in tradeoffs between measurement effort, accuracy, and interpretability . Wepresent Comprex builds similarly accurate performance-influential models to the most accurate and expensive black-boxes approach, but at a reduced cost .…

Mobile edgecomputing (MEC) provides IoT mobile devices with powerful externalcomputing and storage resources . However, a mechanism addressing distributed task offloading and price competition for the open exchange marketplace has not been established properly . We propose two algorithms, namely iterativeproximal offloading algorithm (IPOA) and iterative Stackelberg game pricing algorithm (ISPA) IPOA solves follower non-cooperative game among IoTMDs and ISPA uses backward induction to deal with the price competition amongOSPs .…

We show how to construct and decode liftedinterleaved linearized Reed-Solomon codes . The proposed decoder is a list decoder that can also beinterpreted as a probabilistic unique decoder . We present a heuristic argument and simulation results that indicate that the list size is in fact one for most channelrealizations up to the maximal decoding radius.…

Deep learning has become the gold standard for image processing over the past decade . But two key challenges currently pose amajor barrier to the use of deep learning for vision-based on-orbit proximityoperations . A scarcity of labeled training data (images of a target spacecraft) hinders creation of robust deep learning models .…

Target speech extraction, which extracts the speech of a target speaker in amixture given auxiliary speaker clues, has recently received increased interest . Various clues have been investigated such as pre-recorded enrollmentutterances, direction information, or video of the target speaker .…

In theory, the performance of meta-heuristic search methods is highly dependent on their hyper-parameters . In this paper, we propose a new metric, which estimates how cost-effective tuning a particular classis . We then predict “Tuning Gain” using static features of source code classes .…

Network appliances continue to offer novel opportunities to offload processing from computing nodes directly into the data plane . ZipLine is an approach to design and implement (de)compression at linespeed leveraging the Tofino hardware platform which is programmable using theP4_16 language .…

In this paper, an enhanced Virtual Element Method (VEM) formulation is proposed for plane elasticity . It is based on the improvement of the strainrepresentation within the element, without altering the degree of the interpolating functions on the element boundary .…

The main purpose of this article is to develop a novel refinement strategy for four-dimensional hybrid meshes based on cubic pyramids . This optimal refinement strategy subdivides a given cubic pyramid into a conforming set ofcongruent cubic pyramid and invariant bipentatopes .…

Given a monoid $M$ it is natural to ask how long a sequence of elements needs to be to ensure the presence of consecutive idempotent factors . We study the behaviour of the Ramsey function $R_M$ by studying the regular $D$-length of the monoid .…

An in-depth, longitudinal case study of information systems development in a large manufacturing company was used to investigate howvarious project stakeholders subjectively perceived the project outcome . A conceptual framework is developed for understanding and analyzing evaluations of project success, bothformal and informal .…

This paper concerns the numerical procedure for solving hybrid optimal control problems with sliding modes . The procedure uses the discretization of system equations by Radau IIA Runge–Kutta scheme and the evaluation of optimization functionsgradients with the help of the adjoint equations stated for discretized systemequations .…

Time-dependent wave equations represent an important class of partialdifferential equations (PDE) for describing wave propagation phenomena . We present amachine-learning method to solve this equation as an alternative to ABCs . The mapping from the initial conditions to the PDE solution is represented by a neural network, trained using wave packets that areparameterized by their band width and wave numbers .…

Paper proposes lightweight phone-based transducer model with tiny decoding graph on edge devices . SVD technology compresses the model further. WFST-baseddecoding graph takes the context-independent (CI) phone posteriors as input and allows us to flexibly bias user-specific information. With only 0.9Mparameters after SVD, our system could give a relative 9.1% – 20.5% improvement.…

We conducted a set of Rapid Reviews to characterize Internet of Thingsfacets . We formatted a generic meta-protocol that was instantiated for each of the six facets presented (Connectivity, Things, Behavior, Smartness and Smartness) and considering the issue of Security, one of the most important and frequent challenges in the context of IoT .…

Augmented Informative Cooperative Perception (AICP) is the first fast-filteringsystem which optimizes the informativeness of shared data at vehicles . AICP displays the filtered data to the drivers in augmented reality head-up display . The prototype realizes the informative-optimized perception with only 12.6 milliseconds additional latency .…

Wearable Cognitive Assistance (WCA) amplifies human cognition in real timethrough a wearable device and low-latency wireless access to edge computing infrastructure . Ajalon is a toolchain that reduces the skill and effort needed at each step of the development pipeline .…

In order for agents trained by deep reinforcement learning to work alongside humans in realistic settings, we will need to ensure that the agents are\emph{robust}. Since the real world is very diverse, and human behavior oftenchanges in response to agent deployment, the agent will likely encounter novelsituations that have never been seen during training .…

This note describes the full approximation storage (FAS) multigrid scheme for an easy one-dimensional nonlinear boundary value problem . We apply both FAS V-cyclesand F-cycles, with a nonlinear Gauss-Seidel smoother, to solve the resulting resulting problem . The mathematics of the FAS restriction andprolongation operators are explained .…

A sufficient amount of annotated data is required to fine-tune language models for downstream tasks . Attaining labeled data can be costly, especially for multiple language varieties/dialects . We propose to self-train pre-trained language models in zero- and few-shot scenarios toimprove the performance on data-scarce dialects using only resources from data-rich ones .…

The resulting synthetic Lombard speech has a significant positive impact on intelligibility gain . We alsoshow that the resulting synthetic . speech . has a positive effect on the . resulting . impact on speech . We can generate high quality speech through the .pre-training/fine-tuning…

Failure modes and effects analysis (FMEA) is one of the most practical designtools implemented in the product design to analyze the possible failures and toimprove the design . The use of FMEA is diversified, and different approaches are proposed by various organizations and researchers from one application to another .…

This paper proposes an off-grid channel estimation scheme for orthogonaltime-frequency space (OTFS) systems adopting the sparse Bayesian learning (SBL) framework . To avoid channel spreading caused by fractional delay and Doppler shifts and to fully exploit the channel sparsity in the delay-Doppler(DD) domain, we estimate the original DD domain channel response rather thanthe effective DD channel response as commonly adopted in the literature .…

The timed position of documents retrieved by learning to rank models can be seen as signals . Signals carry useful information such as drop or rise of documents over time or user behaviors . We propose to use thelogic formalism called Signal Temporal Logic to characterize documentbehaviors in ranking accordingly to the specified formulas .…

This work presents a non-intrusive surrogate modeling scheme based on machinelearning technology for predictive modeling of complex systems . The proposed method utilizes a convolutional autoencoder inconjunction with a feed forward neural network to establish a low-cost and accurate mapping from the problem’s parametric space to its solution space .…

Heart Sound (also known as phonocardiogram (PCG) analysis) analysis is a popular way to detect cardiovascular diseases (CVDs) Most PCG analysis uses supervised way, which demands both normal and abnormal samples . This paper proposes amethod of unsupervised PCG . analysis that uses beta variational auto-encoder ($\beta-\text{VAE) to model the normal PCG signals .…

This article introduces a multiple classifier method to improve the performance of concatenate-designed neural networks, such as ResNet and DenseNet . We use the L2 normalization method to obtain the classifier score instead of the Softmax normalization . The proposed classifiers are able to improve accuracy in the experimental cases significantly, and show that the method has better performance than the original models, but also produces faster convergence .…

Using Intelligent Reflecting Surfaces (IRS) can enhance the computing capability of a wireless network scenario . The results confirm the usefulness of IRS technology for future networks — such as 6G — with massive computation requirements, authors say . They say IRS technology can be useful for future wireless networks such as the likes of 6G .…

A paper presents anempirical study of automatic stress-tracking technologies in use in China, based on semi-structured interviews with 17 users . The study highlights three challenges that prevent effective technology usage . The lack of immediate awareness, the lack of pre-required knowledge, and the .…

Can a computer determine a piano player’s skill level? Is it preferable to visual analysis of the player’s performance or should we trust our ears over our eyes? Since current CNNs have difficulty processing long video videos, how can shorter clips be sampled to best reflect the playersskill level?…

A promising feature of passivity theory,alongside stability, has traditionally been claimed to be intuitive performancetuning along the execution of a given task . We introduce optimal energy shaping as an enhancement of classicalpassivity-based control methods . Once a task-dependent performance metric is defined, an optimalsolution is systematically obtained through an iterative procedure relying onneural networks and gradient-based optimization .…

In this paper we investigate confidenceestimation for end-to-end automatic speech recognition (ASR) We provide an extensive benchmark of popular confidence methods on four well-known speech datasets . There are two challenges we overcome in adapting existing methods: working on structured data(sequences) and obtaining confidences at a coarser level than the predictions .…

Nature-inspired optimization can mitigatethis problem by producing compelling yet straightforward solutions when dealing with complicated fitness functions . New mathematicalrepresentations, such as quaternions and octonions, are being used to handle higher-dimensional spaces . The intended hypercomplex feature selection is tested for several meta-heuristic algorithms and hypercomplex representations, achieving results comparable to somestate-of-the-art approaches .…

Most network embedding techniques model the proximity between nodes in a network . Recently there has been significant interest in structural embeddings that are based on node equivalences . They are the first to contribute rigorous intrinsic and extrinsic evaluation methodology .…

This paper studies a class of stochastic and time-varying Gaussianintersymbol interference (ISI) channels . The $i^{th$ channel tap during timeslot $t$ is uniformly distributed over an interval of centre $c_i$ and radius $r_i$. The proposed lower bound is achieved by a joint-typicalityity decoder that is tuned to a set of candidates for the channel matrix .…

We are concerned with random ordinary differential equations (RODEs) Ourmain question of interest is how uncertainties in system parameters propagatethrough the possibly highly nonlinear dynamical system and affect the system’sbifurcation behavior . We come up with a methodology to determine theprobability of the occurrence of different types of bifurcations based on the probability distribution of the input parameters .…

Audio Speech Recognition (ASR) systems are evaluated using Word Error Rate(WER) which is calculated by comparing the number of errors between the groundtruth and the ASR system’s transcription . This calculation, however, requires manual transcription of the speech signal to obtain the ground truth .…

The cloud’s flexibility and promise of seamless auto-scaling notwithstanding, the ability to meet service level objectives (SLOs) typically calls for some form of control in resource usage . This paper investigates that trade-off through the design and implementation of a real-time messaging system motivated by Internet-of-Things(IoT) applications .…

In this work we investigate a coupled system of degenerate and nonlinearpartial differential equations governing the transport of reactive solutes ingroundwater . We show that the system admits a unique weak solution provided thenonlinear adsorption isotherm associated with the reaction process satisfies physically reasonable structural conditions .…

Cyber-Physical Systems (CPSs) combine software and physical components . Thesesystems are widely applied in society within many domains, including theautomotive, aerospace, railway, etc. Testing these systems is extremelychallenging . A driving CPS testing technique in industry issimulation-based testing . However, this poses significant challenges .…

MLGO is the first integration of ML in a complex compiler pass in a real-world setting . We use two different ML algorithms:Policy Gradient and Evolution Strategies, to train the inlining-for-size model . We achieve up to 7\% size reduction, when compared to state of the art LLVM-Oz.…

The proposed method models sceneillumination via a novel, parameterized virtual light stage, which introduces a coarse-to-fineoptimization formulation for face reconstruction . Our method can not only handle unconstrained illumination and self-shadows conditions, but also estimates diffuse and specular albedos . With consistent faceattributes reconstruction, our method leads to several style — illumination,albedo, self-shadow — edit and transfer applications, as discussed in thepaper .…

Personalized gamification explores knowledge about the users to tailorgamification designs to improve one-size-fits-all gamification . The tailoring process should simultaneously consider user and contextual characteristics, which leads to several instances to tailor . The main implications are that demographics, game-related characteristics, geographic location, and geographic location to be done, should be considered in defining gamification designs, as well as the interaction between different kinds of information (user and contextualcharacteristics) can be considered .…

Edge intelligence requires to fast access distributed data samples generated by edge devices . The challenge is using limited radio resource to acquire massive data samples for training machine learning models at edge server . In this article, we propose a new communication-efficient edge intelligence schemewhere the most useful data samples are selected to train the model .…