Transfer Learning for Human & AI

In this paper, we explore how different representations of network trafficaffect the performance of machine learning models for a range of networkmanagement tasks . We study the relationship between the systems-level costs ofdifferent representations of traffic to the ultimate target performancemetric — e.g.,…

A GPU assistedPreemptive MACO (GMACO-P) algorithm has been proposed to minimize the total travel time of the commuters . Inefficient traffic signals and routing mechanisms are the major factors that contribute to the increase of road congestion . For smoother traffic movement and reducingcongestion on the roads, the waiting time at intersections must be reduced and an optimal path should be chosen simultaneously .…

It is known that blood flow shows a non-Newtonian property in small vessels and in situations of complex geometries . In this paper, we focus on investigating the influence on hydrodynamicfactors of different coupled computational models describing the interaction between an incompressible fluid and two symmetric elastic or poroelasticstructures .…

Spiking Neural Networks (SNNs) have the unique ability to detect information encoded in spatio-temporal patterns of spiking signals . Examples include logs of time stamps, e.g., of tweets, and outputs of neuralprostheses and neuromorphic sensors . In this paper, we review models and training algorithms for the dominant approach that considers SNNs as a Recurrent Neural Network (RNN) and adapt learning rules based on backpropagation through time to therequirements of SNN .…

Abstracting Gradual Typing (AGT) is a systematic approach to designinggradually-typed languages . Languages developed using AGT automatically satisfy the formal semantic criteria for gradual languages identified by Siek et al. Nonetheless vanilla AGT semantics can still have importantshortcomings . The default operational semantics of AGT may fail to enforcesome invariants in surprising ways .…

The on-demand mobility market, including ridesharing, is becomingincreasingly important with e-hailing fares growing at a rate of approximately130% per annum since 2013 . The safety of riders and drivers has become of paramount concern and a method for privacy-preserving identity verification between untrusted parties is essential for protecting users .…

Rapid increase of traffic volume on urban roads over time has changed the traffic scenario globally . It has also increased the ratio of road accidentsthat can be severe and fatal in the worst case . There is a need for prediction of severity level of accidents .…

Advances in visual navigation methods have led to intelligent embodiednavigation agents capable of learning meaningful representations from raw RGBimages . However, most learning-based navigation policies are trained and tested in simulation environments . In order for these policies to be useful, they need to be transferred to the real-world .…

Perception for Autonomous Systems (PAZ) is a hierarchical perception library that allow users tomanipulate multiple levels of abstraction in accordance to their requirements . PAZ is divided into three hierarchical levels which we refer to as pipelines, processors, and backends .…

Autonomous vehicles predicted to dominate the transportation industry inthe foreseeable future . Safety is one of the major challenges to the early deployment of self-driving systems . Reits is an antenna system for blindbeamforming that uses constructive interference to return an enhanced radarsignal in the direction of the radar emitter .…

Writing parallel codes is difficult and exhibits a trade-off between abstraction and performance . The high level language abstractions make certain assumptionsthat impacts performance and scalability . On the other hand lower levellanguages, providing many opportunities for optimisation, require in-depth knowledge of parallelism .…

In this paper, we propose a new neural architecture search (NAS) problem ofSymmetric Positive Definite (SPD) manifold networks . Unlike the conventionalNAS problem, our problem requires to search for a unique computational cell called the SPD cell . This SPD cell serves as a basic building block of SPDneural architectures .…

A study is presented on the computation of coupledadvection-diffusion-reaction equations by Schwarz waveform relaxation methods . An optimal interface condition for Schwarzwaveform relaxation is also obtained, which leads to “perfect convergence”, that is, convergence within two times of iteration . The methodsand analyses are extended to the coupling of the viscous Burgers equations .…

Robotics and related technologies are central to the ongoing digitization and advancement of manufacturing . In recent years, a variety of strategic initiatives around the world including “Industry 4.0”, introduced in Germany in 2011 have aimed to improve and connect manufacturing technologies in order to optimize production processes .…

Reinforcement learning has been applied to a wide variety of robotics problems, but most of such applications involve collecting data from scratch . We show that even when the priordata does not succeed at solving the new task, it can still beutilized for learning a better policy, by providing the agent with a broader understanding of the mechanics of its environment .…

A novel neural network architecture can directly generate the cost-to-go (c2g) function for a given configuration space and a goal configuration . The output of the network is acontinuous function whose gradient in configuration space can be directly used to generate trajectories in motion planning without the need for protractediterations or extensive collision checking .…

Reinforcement Learning (RL) is widely utilized in the field of robotics, and as such, it is gradually being implemented in the Hybrid Electric Vehicle (HEV)supervisory control . This study aims to reduce the learning iterationsof Q-learning in HEV application and improve fuel consumption in initiallearning phases .…

Sampling-based planning methods often become inefficient due to narrow passages . Narrow passages induce a higher runtime, because the chance to sample them becomes vanishingly small . Solving the relaxed problem is oftenefficient and produces an admissible heuristic we can exploit .…

MuJoCo is the best engine to transfer the learning to other engines . None of the algorithms generalize when trained on PyBullet. We also found out thatvarious algorithms have a promising generalizability if the effect of randomseeds can be minimized on their performance.…

Combining multiple LiDARs enables a robot to maximize its perceptualawareness of environments and obtain sufficient measurements . This paper proposes a system to achieve robust and simultaneous extrinsic calibration, odometry,and mapping . The source code,datasets, and demonstrations are available athttps://ram-lab.com/file/site/m-loam .…

CopyPaste is a perceptually motivated novelaugmentation procedure for SER . It suggests that the presence of emotions otherthan neutral dictates a speaker’s overall perceived emotion in a recording . We hypothesize that SER performance can be improvedusing these concatenated utterances in model training .…

Deep speaker embeddings have become the leading method for encoding speakeridentity in speaker recognition tasks . The embedding space should ideally capture the variations between all possible speakers, encoding the multiple aspects that make up speaker identity . We show that by leveraging additional speaker attributeinformation in a multi task learning setting, deep speaker embeddingperformance can be increased for verification and diarization tasks, achieving a relative improvement of 17.8% in DER and 8.9% in EER for Supreme Court audiocompared to omitting the auxiliary task .…

End-to-end (E2E) automatic speech recognition (ASR) models, by now, haveshown competitive performance on several benchmarks . This workpresents cascaded encoders for building a single E2E ASR model that can operate in both modes simultaneously . Results show that this model achieves similar word error rates (WER) as a standalone streaming modelwhen operating in streaming mode, and obtains 10% — 27% relative improvement when operating in non-streaming mode .…

For the first time, we develop a convergent numerical method for the llinearintegral equation derived by M.M. Lavrent’ev in 1964 with the goal to solve acoefficient inverse problem for a wave-like equation in 3D . The data are nonoverdetermined. Nevertheless, numerical results forthat equation, which use the data generated for that coefficient inverseproblem, show a good reconstruction accuracy .…

This paper studies numerical solutions for parameterized partial differentialequations (P-PDEs) with deep learning (DL) P-PDE arise in many important application areas and computational cost using traditional numerical schemes can be exorbitant . If we directly apply existing supervised learning models, the models need to be constantly fine-tuned or retrained when the parameters change .…

Global Navigation Satellite System(GNSS) is getting increasing attention for effective use of its timedissemination to vehicular networks . In urban road scenarios, blockage of satellite signals affects precise positing due to erroneous geometry . With the recent development of Multi-GNSS receivers, signal receptionno longer relies on a single GNSS constellation but all the existing four GNSSsystems .…

A number of recent advances in audio synthesis rely on neural upsamplers, which can introduce undesired artifacts . In computer vision, upsampling artifacts have been studied and are known as checkerboard artifacts . Their effect has been overlooked so far in audio processing.…

This letter introduces HOPPY, an open-source, low-cost, robust, and modularkit for robotics education . The robot dynamically hops around a rotating gantry with a fixed base . The kit bridges the theoretical content of fundamental robotic courses and real dynamic robots by facilitating and guiding the software and hardware integration .…

Traditional methods on excavationtrajectory generation usually separate the excavation motion into a sequence offixed phases, resulting in limited trajectory searching space . Our frameworkexplores the space of all possible excavation trajectories represented with waypoints interpolated by a polynomial spline . We formulate a generic task specification forexcavation by constraining the instantaneous motion of the bucket and furtheradd a target-oriented constraint, i.e.…

The randomized Kaczmarz algorithm has received considerable attention recently because of its simplicity, speed, and the ability to approximately solve large-scale linear systems of equations . The proposed algorithms avoidforming $A^\top A$ explicitly and work for $m\geq n$ (full rank or rank deficient) Upper bounds showing exponential convergence in the mean square sense of the proposed algorithms are presented and numerical experiments are given to illustrate the theoretical results .…

Transporter Network is asimple model architecture that rearranges deep features to infer spatialdisplacements from visual input – which can parameterize robot actions . It is orders of magnitude more efficient than our benchmarked alternatives in learning vision-basedmanipulation tasks . Our method can represent complex multi-modalpolicy distributions and generalizes to multi-step sequential tasks, as well as 6DoF pick-and-place .…

This paper proposes voicing-aware conditional discriminators for ParallelWaveGAN-based waveform synthesis systems . We adopt aprojection-based conditioning method that can significantly improve thediscriminator’s performance . As each discriminator learns the distinctive characteristics of theharmonic and noise components, the adversarial training process becomes more efficient, allowing the generator to produce more realistic speechwaveforms .…

We consider the optimization of the vector of grasping forces that support a known generalized force acting on the grasped object . We present an expression for the optimum which depends on the external force and the kinematics of the grasping system .…

ByteCover is built based on the classicalResNet model, and two major improvements are designed to further enhance the model for CSI . ByteCover outperformed the best competitive system by 20.9\% . A set of experiments demonstrated the effectiveness and efficiency of ByteCover on multiple datasets, and in theDa-TACOS dataset, ByteCover outranks the best competition system by 20.9 \% .…

The solution of linear systems of equations is a central task in a number of scientific and engineering applications . We show how adaptive FSAI, an approximate inverse characterized by a very high degree of parallelism, can be successfully implemented on a distributed memory computer equipped with GPUaccelerators .…

We consider acoustic scattering in heterogeneous media with piecewiseconstant wave number . The discretization is carried out using a Galerkinboundary element method in space . We prove well-posedness of the scheme and provide a priori estimates for the convergence in space and time .…

Real-world LoRaWAN datasets are important for repeatable sensor-network and communications research and evaluation . Data is collected from nine gateways over a four month period in a dense urban environment . The dataset contains packet header information and all physicallayer properties reported by gateways such as the CRC, RSSI, SNR and spreadingfactor .…

In this paper, we empirically examine evidence of recurring dependency management issues (dependency smells) We look at commit data for a dataset of 1,146 active JavaScript repositories . Practitioners agree that dependency smells bring about many problems including security threats, bugs,dependency breakage, runtime errors, and other maintenance issues .…

Several applications in optimization, image and signal processing deal with the Stiefel manifold St(n, p), that is, the set of n x pmatrices with orthonormal columns . Some applications, like the Karcher mean, require evaluating the geodesic distance between two arbitrary points .…

Various information factors are blended in speech signals, which forms the difficulty for most speech information processing tasks . An intuitiveidea is to factorize speech signal into individual information factors (e.g.,phonetic content and speaker trait), though it turns out to be highlychallenging .…

System achieves an equal error rate of 2.19% on the evaluation set of ASVspoof 2019 Challenge, outperforming all existing single systems . The key idea is tocompact the genuine speech representation and inject an angular margin to separate the spoofing attacks in the embedding space of the speech .…

In this study, we present recent developments on ESPnet: End-to-End SpeechProcessing toolkit, which mainly involves a recently proposed architecture called Conformer, Convolution-augmented Transformer . Our experiments reveal various training tips and significant performance benefits obtained with the Conformer on different tasks .…

Online end-to-end ASR still lags far behind that of offline systems . For specific scenarios, we cantrade-off between performance and latency . We propose a novel architecture, termed Universal ASR, that can unify streaming and non-streaming ASR models into one system .…

Domain mismatch often occurs in real applications and causes seriousperformance reduction on speaker verification systems . We present a decoupledscoring approach that can maximally squeeze the value of cross-domain labels when the enrollment and test aremismatched . When the statistics are coherent, the new formulation falls back to the conventional PLDA.…

The star chromatic index is the minimum number of colors needed to properlycolor the edges of $G$ such that no path or cycle of length $4$ is bicolored . In this paper, we study the star edge-coloring of Halin graphs, $k$-powergraphs and the generalized Petersen graphs $P(3n, n)$.…

A word is said to be bordered if it contains a non-empty proper prefix that is also a suffix . We give a recurrence for the number of mutually bordered words of length-$n$ over a $k$-letter alphabet . We show that, asymptotically, there are $c\cdotk^{2n}$ mutually-bordered words over a letter alphabet, where $c$ is a constant .…

In this manuscript we consider the set of Dyck paths equipped with the uniform measure . We study the statistical properties of a deformation of the observable “area below the Dyck path” as the size $N$ of the path goes toinfinity .…

We propose a one parameter family of random walk processes on hypergraphs . A parameter biases the dynamics of the walker towards hyperedges of low or high cardinality . We show that for each value of the parameter the resulting process defines its own hypergraph projection on a weighted network .…

Graph Neural Networks (GNNs) are the predominant technique for learning overgraphs . However, there is relatively little understanding of why GNNs are necessary for good performance . Here, we show that for many standard transductive node classificationbenchmarks, we can exceed or match the performance of state-of-the-art GNN models by combining shallow models that ignore the graph structure with two simple post-processing steps that exploit correlation in the label structure .…

Quantum topological invariants have played an important role in computationaltopology . They are at the heart of major modern mathematical conjectures . We study the experimental problem of computing large $r$values of Turaev-Viro invariants . We base our approach on anoptimized backtracking algorithm, consisting of enumerating combinatorial data on a triangulation of a 3-manifold .…