Transfer Learning for Human & AI

We present three families of minimal border rank tensors: they come from highest weight vectors, smoothable algebras, or monomial algesbras . We analyseethem using Strassen’s laser method and obtain an upper bound $2.431$ on$\omega$. We also explain how in certain monomial cases using the laser methoddirectly is less profitable than first degenerating .…

We consider a dynamic Colonel Blotto game (CBG) in which one of the players has limited troops (budget) to allocate over a finite time horizon . At each stage, the learner strategically determines the budget and its distribution to allocate among the battlefields based on past observations .…

We introduce a simple, accurate, and extremely efficient method fornumerically solving the multi-marginal optimal transport (MMOT) problems . The method relies on (i) the sparsity ofoptimal plans [for $N$ marginals discretized by $\ell$ gridpoints each] and ideas from machine learning .…

Phishing is one of the most severe cyber-attacks where researchers are trying to find a solution . In phishing, attackers lure end-users and steal their personal in-formation . There are various phishing attacks like spearphishing, whaling, vishing, smishing, pharming and so on .…

ESCORT leverages a multi-output NNarchitecture that consists of two parts: (i) A common feature extractor that learns the semantics of the input contract; (ii) Multiple branch structures where each branch learns a specific vulnerability type based on features . ESCORT achieves an average F1-score of 95% on six vulnerability types and the detection time is 0.02 seconds per contract .…

In this paper, we study linear filters to process signals defined onsimplicial complexes . We propose a finite impulse response filter based on the HodgeLaplacian . We demonstrate how this filter can be designed to amplify orattenuate certain spectral components of simplicial signals .…

This paper describes an approach to solving the next destination city recommendation problem for a travel reservation system . We propose a two stagesapproach: a heuristic approach for candidates selection and an attention neuralnetwork model for candidates re-ranking . We used this approach to solve the Booking.com…

Contact tracing via digital tracking applications installed on mobile phones is an important tool for controlling epidemic spreading . Our results are based on a combination of explicitsimulations and mean-field analysis . They indicate that there can be majordifferences in the epidemic size and epidemic probabilities which are equivalent in the normal SIR processes.…

Biometric authentication technology for smartphones has become widespread . However, fingerprint authentication cannot be used when a person is wet or wearing amask . We examine a personal authentication system using the pinna as a new approach for biometric authentication on smartphones .…

This paper is concerned with the design of a non-intrusive model orderreduction (MOR) for the system of parametric time-domain Maxwell equations . Atime- and parameter-independent reduced basis (RB) is constructed by using atwo-step proper orthogonal decomposition (POD) technique from a collection offull-order electromagnetic field solutions, which are generated via adiscontinuous Galerkin time- domain (DGTD) solver .…

Blow-up solutions of autonomous ordinary differentialequations (ODEs) which are unstable under perturbations of initial conditions are studied . They are obtained as trajectories on stable manifolds of hyperbolic (saddle) equilibria atinfinity . In this process, important features are obtained: smooth dependenceof blow-up times on initial points near blow-ups, level set distribution ofblow-upsolutions, singular behavior of blow- upsolutions, organization of the phase space via separatrices (stable manifolds) In particular, we show that unstable blow up solutions themselves canseparate initial conditions into two regions where solution trajectories are globally bounded or blow up, no matter how large initial points are…

We present an open-source space-and doubly-time-parallel implementation and evaluate its performance for twodifferent test problems . We use a collocationmethod of arbitrary order to add another level of parallelization in time . We derive an adaptive strategy to select a new $alpha$-circulantpreconditioner for each iteration during runtime for balancing convergencerates, round-off errors and inexactness in the individual time-steps .…

Many multilinear maps have many cryptographic applications, but secure and efficient construction of such maps is an open problem . Attempt is made to propose a new GES, where, instead of encoding an element, users can obtain the encoding of an associated but unknown random element .…

Artificial Neural Networks (ANNs) became popular due to their successfulapplication difficult problems such as image and speech recognition . We propose a framework called AutoLR to automatically design Learning RateOptimizers . The results are competitive with the best state of the art methods, even outperforming them in some scenarios .…

This work considers {\emuniform global time-renormalization functions for the {\em gravitational}$N$-body problem . It improves on the estimates of the radii of convergence by using a completely different technique . The technique which the new estimates are built upon is known as {\em majorants .…

Open Source Software projects add labels to open issues to help contributors choose tasks . Current automatic approaches for creating labels are mostly limited to classifying issues as a bug/non-bug . We leverage the issues’ description and the project history to build prediction models, which resulted in precision up to82% and recall up to 97.8% .…

Deep learning models for forecasting resource consumption patterns of queries have recently been a popular area of study . But training them over large scale industry workloads are expensive . In turn, we developed Prestroid, atree convolution based data science pipeline that accurately predicts resourceconsumption patterns of query traces .…

We address the issue of point value reconstructions from cell averages in the context of third order finite volume schemes . In fact, most techniques known inthe literature rely on the creation of ghost cells outside the boundary and on extrapolation from the inside that, taking into account theboundary conditions, fills the ghost cells with appropriate values .…

The study examined the participation of female students of South EasternNigerian tertiary institutions in Information and Communication Technologies(ICTs) The study discussed the attendant gender divide in ICTs participation . Findings suggest that high cost of ICT and highlevel of male dominance, which made females think that ICT is for males were the major reasons for low female participation .…

In this paper, we study secure energy efficiency (SEE) for simultaneouswireless information and power transfer (SWIPT) in a distributed antenna system(DAS) based IoT network . We consider a system in which both legitimate users(Bobs) and eavesdroppers (Eves) have power splitting (PS) receivers to decode information and harvest energy from the received signal .…

Developing complex engineered systems poses significant challenges forengineers, managers, designers, and businesspeople alike . Experts have expressed great interest in filling the gap in theory about how CES develop . The ComplEx System IntegratedUtilities Model (CESIUM) is a novel framework for exploring how numerous systemand development process characteristics may affect the performance of CES .…

Money laundering (ML) is the behavior to conceal the source of money achievedby illegitimate activities . CubeFlow is a scalable, flow-based approach to spot fraud from a mass of transactions by modeling them as two coupled tensors and applying a novelmulti-attribute metric which can reveal the transfer chains accurately .…

Program translation is a growing demand in software engineering . Manualprogram translation requires programming expertise in source and targetlanguage . One way to automate this process is to make use of the big data ofprograms, i.e., Big Code . We propose a lightweight but informative programrepresentation, which can be generalized to all imperative PLs .…

We devise ageneral method called Multi-Level Splitting Sampling (MLSS) that can handle complex queries and complex models . Our method addresses the inefficiency of standard Monte Carlo (MC) methods by applying the idea of importance splitting to let one “promising” sample path prefix generate multiple “offspring” paths .…

Hybrid complex analytics workloads typically include (i) data managementtasks (joins, selections, etc. ), easily expressed using relational algebra(RA)-based languages, and (ii) complex analytics tasks (regressions, matrixdecompositions) Such workloads are common in many application areas, including scientificcomputing, web analytics, and business recommendation .…

Understanding how beliefs are accepted, spread, and intensified is critical for any attempt to mitigate radicalization and avoid increased political polarization . This article presents an agent-based model of a social media network that enables investigation of the effects of censorship on the amount of dissenting information to which agents become exposed and thecertainty of their radicalized views .…

Neuromorphic computing systems such as DYNAPs and Loihi have recently been introduced to the computing community to improve performance and energyefficiency of machine learning programs . The role of a system software forneuromorphic systems is to cluster a large machine learning model (e.g.,…

In this paper we investigate the problem of controlling a partially observed dynamical system such that its state is difficult to infer using a(fixed-interval) Bayesian smoother . This problem arises naturally in applications in which it is desirable to keep the entire state trajectory of asystem concealed .…

An Infrastructure as a Service (IaaS) cloud provider is committed to eachtenant by a service level agreement (SLA) This may violate the SLAs towards thetenants and result in penalty . The upgrade of IaaS cloud systems inherits all the challenges of clustered systems and faces other cloud specific challenges, such as size and dynamicitydue to elasticity .…

Real-world graphs often have an immense size, such that loading the complete graph into memory for partitioning is not economical or feasible . Hybrid Edge Partitioner (HEP) can partition graphs that fit partly intomemory while yielding a high partitioning quality .…

Many online platforms are providing valuable real-time contents (e.g.,traffic) by continuously acquiring the status of different Points of Interest(PoIs) In status acquisition, it is challenging to determine how frequently aPoI should upload its status to a platform, since they are self-interested with private and possibly time-varying preferences .…

Low-order finite-element discretizations are well-known to provide effectivepreconditioners for linear systems that arise from higher-orderdiscretizations of the Poisson equation . In this work, we show thathigh-quality preconditioners can also be derived for the Taylor-Hooddiscretization of the Stokes equations in much the same manner .…

Multigrid solvers for hierarchical hybrid grids (HHG) have been proposed topromote the efficient utilization of high performance computer architectures . HHG meshes are constructed by uniformly refining a relatively coarsefully unstructured mesh . This paper focuses ongeneralizing the HHG idea so that it is applicable to a broader community ofcomputational scientists .…

In this review, we describe the application of one of the most popular deeplearning-based language models – BERT . The paper describes the mechanism ofoperation of this model, the main areas of its application to the tasks of textanalytics, comparisons with similar models in each task, as well as adescriptions of some proprietary models .…

Machine learning (ML) is successful in achieving human-level performance invarious fields . However, it lacks the ability to explain an outcome due to its black-box nature . We propose a novel framework for accelerating explainable ML using Tensor Processing Units (TPUs) The proposed framework exploits the synergy between matrix convolution and Fourier transform, and takes full advantage of TPU’s natural ability in accelerating matrix computations .…

We comment on two formal proofs of Fermat’s sum of two squares theorem . The first one follows Zagier’s celebrated one-sentence proof; the second relies on partition-theoretic arguments . Both formal proofs rely on a general property of involutions offinite sets, of independent interest .…

We propose a functional implementation of \emph{Multivariate Tower AutomaticDifferentiation}. Our implementation is intended to be used in implementing$C^\infty$-structure computation of an arbitrary Weil algebra .…

The best performing technique, Longformer, achieved an average F1 score of 80.99% (F1=99.68% for SpinBot and F1=71.64% for SpinnerChief cases) We show that the automated classificationalleviates shortcomings of widely-used text-matching systems, such as Turnitinand PlagScan . To facilitate future research, all data, code, and two webapplications showcasing our contributions are openly available .…

The technique involves coupling and manipulating two chaotic systems to minimize the coveragetime and enable scanning of unknown environments with different properties online . Using this technique resulted in 49% boost, on average, in the robot’s performance compared to the state-of-the-art planners .…

Business process simulation is a well-known approach to estimate the impact of changes to a process with respect to time and cost measures . Data-Driven Simulation (DDS)methods combine automated process discovery and enhancement techniques to learnprocess simulation models from event logs .…

The incompressible Navier-Stokes equations are solved in a channel, using aDiscontinuous Galerkin method over staggered grids . The resulting linearsystems are studied both in terms of the structure and in terms the spectralfeatures of the related coefficient matrices . In fact, the resulting matricesare of block type, each block showing Toeplitz-like, band, and tensor structure at the same time .…

Standard automatic metrics (such as BLEU) are problematic for document-levelMT evaluation . To address these problems, we propose an automatic metric BlonD . It takes discourse coherence into consideration by calculating the recall and distance of check-pointing phrases and tags, and provides comprehensive evaluation scores by combining with n-gram .…

Distributed training across several quantum computers could significantlyimprove the training time . If we could share the learned model, not the data, it could potentially improve the data privacy as the training would happen where the data is located . However, to the best of our knowledge, nowork has been done in quantum machine learning (QML) in federation setting yet .…

Feature Selection (FS) offers several advantages, s.a. decreasing computational costs, aiding inference and interpretability . Traditional FS techniques may become sub-optimal in the presence of imbalanced data . We propose a filtering FS algorithm ranking feature importance on the basis of theReconstruction Error of a Deep Sparse AutoEncoders Ensemble (DSAEE) We use each DSAE trained only on majority class to reconstruct both classes .…

Algorithmic decision-making (ADM) increasingly shapes people’s daily lives . We provide a comprehensive,systematic literature review synthesizing the existing empirical insights on the perception of algorithmic fairness . We hope our work will contribute to a society-in-the-loop framework, we say . By advocating for more interdisciplinary research, we hope to contribute to fairer and more responsible ADM.…

We address the problem of performing backpropagation for computation graphs involving 3D transformation groups . 3Dtransformation groups are widely used in 3D vision and robotics, but they donot form vector spaces and instead lie on smooth manifolds . We show that our approach is numericallymore stable, easier to implement, and beneficial to a diverse set of tasks .…

This study presents an approach to lane detection involving the prediction of binary segmentation masks and per-pixel affinity fields . These affinity fields,along with the binary masks, can then be used to cluster lane pixelshorizontally and vertically into corresponding lane instances in apost-processing step .…

New software FEniCS-preCICE is a middle software layer, sitting inbetween the existing finite-element library FeniCS and the coupling library preCICE . The middle layer simplifies coupling (existing) applicationcodes to other simulation software . Only a few lines of additional code are necessary to prepare a code for coupling .…

The analysis of the compression effects in generative adversarial networks after training remains an unstudied topic . High compression levels may distort the generated set, likely leading to an increase of outliers that may negatively affect the overall assessment of existing k-nearest neighbor (KNN)-based metrics .…