Transfer Learning for Human & AI

This paper describes the APT (AutomatedProgram Transformations) tools to carry out isomorphic data type refinements . The tools are also useful to verify existing programs, by turning more concrete data representations into more abstract ones to ease verification . Propagate-iso handles cases where the type is a component of a more complex one such as a list of the type or a record that has a field is automatically lifted to an isomorphism on the more complex type .…

Recursive Length Prefix (RLP) is used to encode a wide variety of data inEthereum, including transactions . The work described in this paper provides a specification of RLP encoding and a verified implementation . This work has led toimprovements to the Ethereum documentation and additions to the testsuite .…

Defret-mutual-generate is a utility for proving ACL2 theorems about large mutually recursive cliques of functions . It builds on previous toolssuch as defret-Mutual and make-flag, which automate parts of the process but still require a theorem body to be written out for each function in the clique .…

RP-Rewriter (Retain-Property) is a verified clause processor that can usesome of the existing ACL2 rewrite rules to prove conjectures through termrewriting . Optimized for conjectures that can expand into large terms, therewriter tries to mimic some of the ACL2 rewriting heuristics but also addssome extra features .…

A common technique for checking properties of complex state machines is tobuild a finite abstraction then check the property on the abstract system . We developed a process for computing a finiteabstract graph of the target state machine . We theniteratively divide the abstract graph by splitting into strongly connected components and then building a measure for every node in the abstractgraph .…

FGL is a successor to GL, a proof procedure for ACL2 that allows complicatedfinitary conjectures to be translated into efficient Boolean function representations and proved using SAT solvers . The FGL rewriter is modeled on ACL2’s rewriter, but we have added several features in order to make rewrite rules more powerful .…

ACL2 algorithms are traditionally expressed in ACL2 using recursion . Common Lisp provides a construct, loop, which provides direct support for iteration . We describe anACL2 analogue loop$ of loop that supports efficient ACL2 programming .…

An intermediate model of a Verilog module is hand-coded in RestrictedAlgorithmic C (RAC) It is then translated to the logic of ACL2, enabling a mechanically checkedproof of correctness . In this paper, we describe the RAC language, the translation process, and sometechniques that facilitate formal analysis of the resulting ACL2 code .…

During an epidemic, infectious individuals might not be detectable until sometime after becoming infected . Carriers with mild or nosymptoms are the main contributors to the transmission of a virus within thepopulation . The average time it takes to develop the symptoms causes a delay inthe spread dynamics of the disease .…

The study of robotic flocking has received considerable attention in the past twenty years . As we begin to deploy flocking control algorithms on physicalmulti-agent and swarm systems, there is an increasing necessity for rigorous promises on safety and performance .…

Q-Mixing is a transfer learning method that starts by learning Q-values against each pure-strategy opponent . Then a Q-value for any distribution of opponents is approximated by averaging the separately learned Q values . From these components, we construct policies against all mixtures without any further training .…

COVID-19 puts real-time (or low-latency) codecs into a new light . It is shown how the speed at which content is made available impacts both latency and throughput . The authors introduce a new test set up, integrating a paced reader, which allows to runcodec in the same condition as real time media capture.…

In the tile-based 360-degree video streaming, predicting user’s futureviewpoints and developing adaptive bitrate (ABR) algorithms are essential foroptimizing user’s quality of experience (QoE) Traditional single-user based prediction methods fail to achieve good performance in long-termprediction . We propose a cross-userattentive network (CUAN) boosting the performance of long-term viewpointprediction .…

Sparse general matrix-matrix multiplication (spGEMM) is an essential component in many scientific and data analytics applications . Modern GPUs include Tensor Core Units (TCUs), which specialize in dense matrix multiplication . Our aim is to re-purpose TCUs forsparse matrices using mixed precision mode of TCUs .…

BOML is a modularized optimization library that unifies several meta-learning algorithms into a common bilevel optimization framework . It provides a hierarchical optimization pipeline together with a variety ofiteration modules . The library is written in Python and is available athttps://://://github.com/dut-media-lab/BOML.…

We introduce a residual-based a posteriori error estimator for a novel$hp$-version interior penalty discontinuous Galerkin method for the biharmonicproblem in two and three dimensions . We prove that the error estimate provides an upper bound and a local lower bound on the error, and that the lower boundis robust to the local mesh size but not the local polynomial degree .…

A heat equation with uncertain domains is thoroughly investigated . A rigorous proof for the existence of the shape derivative is presented . Boundary integral equation methods are used to compute statisticalmoments of shape derivative . The shape derivative has been defined as the shape of the heat equation .…

In this work we propose a novel block preconditioner, labelled ExplicitDecoupling Factor Approximation (EDFA), to accelerate the convergence of Krylovsubspace solvers used to address the sequence of non-symmetric systems of linear equations . The proposed blockpreconditioner has been tested through an extensive experimentation on both synthetic and real-case applications, pointing out its robustness and efficiency .…

The method of the fundamental solutions (MFS) is used to construct an approximate solution for a partial differential equation in a bounded domain . The existence of the approximate solution by the MFS for the Neumannproblems of the modified Helmholtz equation in disk domains is rigorouslydemonstrated .…

One layerneural networks that are $L$-Lipschitz are dense in the set of all$L$ functions with respect to the uniform norm on bounded sets .…

We propose a $p$-multilevel preconditioner for Hybrid High-Orderdiscretizations (HHO) of the Stokes equation . We numerically assess its performance on two variants of the method, and compare with a classical Galerkin scheme . An efficient implementation is proposed wherecoarse level operators are inherited using $L^2$-orthogonal projections defined over mesh faces and the restriction of the fine grid operators is performed recursively and matrix-free .…

A numerical scheme based on modified method of characteristics with adjustedadvection (MMOCAA) is proposed to approximate the solution of the system liquidchromatography with multi components case . For the case of one component, themethod preserves the mass . Various examples and computational tests numericallyverify the accuracy and efficiency of the approach .…

The present paper deals with the interior solid-fluid interaction problem inharmonic regime with randomly perturbed boundaries . Analysis of the shapederivative and shape Hessian of vector- and tensor-valued functions is provided . Moments of the random solutions are approximated by those of the shapes of the shape derivative and shapes Hessian .…

Convolutional neural networks (CNNs) are commonly used for image classification . Saliency methods are examples of approaches that can be used tointerpret CNNs post hoc, identifying the most relevant pixels for a prediction following the gradients flow . We propose a novel approach for training trustworthy CNNs by penalizing parameter choicesthat result in inaccurate saliency maps generated during training .…

Deep facial expression recognition faces two challenges that both stem from the large number of trainable parameters: long training times and a lack ofinterpretability . We propose a novel method based on evolutionary algorithms, that deals with both challenges . We are robustly able to reduce thenumber of parameters on average by 95% (e.g.…

This paper proposes a client selection method for federated learning (FL)when the computation and communication resource of clients cannot be estimated . The method trains a machine learning (ML) model using the rich data andcomputational resources of mobile clients without collecting their data systems .…

Machine-learning-based prediction of future wireless link quality is an emerging technique that can potentially improve the reliability of wireless communications . The proposed system collectsdatasets, updates a model, and infers the received power in real-time . The experimental evaluation was conducted using 5 GHz Wi-Fi, where received signalstrength could be degraded by 10 dB when the LOS path was blocked by largeobstacles.…

Internet Photonic Sensing (IPS) is a new framework for deformation and vibration measurement and monitoring . IPS is based on the hypothesis that atmospheric, seismic,anthropogenic and other natural activity cause vibrations in the earth that trigger detectable changes in standard optical signals that transmit datathrough Internet fiber .…

Flapping wing flight is a challenging dynamical problem and is also a veryfascinating subject to study in the field of biomimetic robotics . Aerobat captures multiple biologically meaningfuldegrees-of-freedom for flapping flight that is present in biological bats . The work attempts to manifest closed-loop aerial body reorientation andpreparation for landing through the manipulation of inertial dynamics andaerodynamics by enforcing nonholonomic constraints onto the system .…

In this paper, we present a planner for a master dual-arm robot to manipulatetethered tools . The assistant robot provides assistance to the master robot by manipulating the tool cable and avoiding collisions . Simulations and real-world experiments are performed tovalidate the proposed planner .…

In silico, data-driven models build, adapt and improve representations of the targetsystem using real-world experimental data . In reality, large-scale physical experimentation facilitates the fabrication,testing and analysis of multiple candidate designs . Automated assembly andreconfigurable modular systems enable significantly higher numbers ofreal-world design evaluations than previously possible .…

In SLAM, researchers usually build and update a3D map after applying non-linear pose graph optimization techniques . Many multi-agent systems are prevalently using odometry information from additional sensors . The proposed framework performs actively in terms oflocalizing each agent after the first loop closure between them .…

This paper considers the problem of parameter identification for a multirobotsystem . We wish to understand when is it feasible for an adversarial observer to reverse-engineer the parameters of tasks being performed by a team of robotsby simply observing their positions .…

Extrinsic perturbation always exists in multiple sensors . MLOD is a two-stage network where the multi-LiDAR information is fused through various schemes in stage one, and the extrinsicperturbation is handled in stage two . We conduct extensive experiments on areal-world dataset, and demonstrate both the accuracy and robustnessimprovement of MLOD .…

Optimal Transport (OT)-based decentralized collaborative multi-robot exploration strategy is proposed in this paper . This method is to achieve an efficient exploration with a predefined priority in the given domain . In thiscontext, the efficiency indicates how a team of robots (agents) cover the domain reflecting the corresponding priority map .…

In this study, an ergodic environment exploration problem is introduced for acentralized multi-agent system . Given the reference distribution represented by the Mixture of Gaussian (MoG), the ergodicity is achieved when thetime-averaged robot distribution is identical to the given referencedistribution .…

Planners using accurate models can be effective for accomplishingmanipulation tasks in the real world, but are typically highly specialized and require significant fine-tuning to be reliable . In this paper, we propose a method that improves efficiency of sub-optimalplanners with approximate but simple and fast models by switching to amodel-free policy when unexpected transitions are observed .…

Digitisation of fruit trees using LiDAR enables analysis which can be used tobetter growing practices to improve yield . Sophisticated analysis requiresgeometric and semantic understanding of the data, including the ability to identify individual trees as well as identifying leafy and structural matter .…

Report describes our submission to the VoxCeleb Speaker RecognitionChallenge (VoxSRC) at Interspeech 2020 . We perform a careful analysis of speaker recognition models based on the popular ResNet architecture . Our results shows significant improvements over most existing works without the use of modelensemble or post-processing .…

In this paper, we introduce a simple method that can separate arbitrary instruments from an audio mixture . Given an unaligned MIDItranscription for a target instrument from an input mixture, we synthesize newmixtures from the midi transcription that sound similar to the mixture to beseparated .…

Heraklit provides means, principles, and unifying techniques to model and to analyze digital infrastructures . Modern and next generation digital infrastructure systems are technically based on service oriented structures, cloud services, and other architectures thatcompose large systems from smaller subsystems .…

We present a new conceptual definition of ‘productivity’ for sustainablydeveloping research software . Existing definitions are flawed as they are short-term biased, thus devaluing long-term impact . We view the outputs of the development process as knowledge and usersatisfaction . User satisfaction is used as a proxy for effective quality .…

This paper describes an in depth analysis of successful and unsuccessful software projects and the role of Software Project Mangers in that success . The average ratio of software project failure is on the high side . Project managers experience, technical knowledge,and skills are not good enough for success in general .…

In recent years, there has been significant progress in the development and adoption of static analyzers . A major hurdle is tuning options to custom usagescenarios, such as a particular code base or certain resource constraints . We propose a technique that automatically tailors a staticanalyzer, specifically an abstract interpreter, to the code under analysis and any given resource constraints.…

NNSlicer is the first approach for slicing deep neural networks based on data flow analysis . The method understands the reaction of each neuron to an input based on the difference between its behavior activated by the input and the average behavior over the whole dataset .…

We propose a totally functional view of geometric matrix completion problem . On synthetic tasks with strong underlying geometric structure, ourframework outperforms state of the art by a huge margin . On real datasets, weachieve state-of-the-art results at a fraction of the computational effort of previous methods .…

Given a combinatorial optimization problem, can we learn a strategy to solve it from the examples of input-solution pairs without knowing its objective function? In this paper, we consider such a setting and study themisinformation prevention problem . We design a structured prediction framework, where the main idea is toparameterize the scoring function using random features constructed through distance functions on randomly sampled subgraphs .…

In many real world datasets arising from social networks, there are hiddenhigher order relations among data points which cannot be captured using graphmodeling . In this paper, we introduce a new local geometry ofhyperdges in hypergraphs which allows to capture higher order relations .…

Hypergraph data appear and are hidden in many places in the modern age . Theyare data structure that can be used to model many real data examples . One of the main contributions of our paper is to introduce a new topological structure to hypergraph data .…

In this whitepaper we advocate that the Planetary Science (PS) communitybuild a discipline-specific digital library . We suggest that the PS data archives increase the level of curation to allow for direct linking between archival data and the derived journal articles .…