Transfer Learning for Human & AI

We propose an Open Source Software Project Evaluation and Selection TOol (OSS PESTO) Targeting OSS on Github, the largest OSS source code host, it facilitates theevaluation practice by enabling practitioners to compare candidates . It also allows in-time Github datacollection and customized evaluation that enriches its effectiveness and easeof use .…

Strapdown inertial navigation research involves the parameterization andcomputation of the attitude, velocity and position of a rigid body in a chosen reference frame . This paper proposes a compact andelegant representation of the body’s attitude, position and velocity . Thekinematics of strapdown INS are cohesively unified in one concise differentialequation .…

Traditional power system frequency dynamics are driven by Newtonian physics,where a synchronous generator (SG), the historical primary source of power,follows a deceleration frequency trajectory upon power imbalances . Subsequent to a disturbance, an SG will modifypre-converter, mechanical power as a function of frequency .…

In machine learning, this approach is often referred to as transferlearning . In this work, we analyze surgical step recognition on four different laparoscopic surgeries: Cholecystectomy, Right Hemicolectomy, Sleeve Gastrectomy, and Appendectomy . We introduce a new architecture, the Time-Series Adaptation Network (TSAN), an architectureoptimized for transfer learning .…

We prove the existence of Bayesian Nash Equilibrium (BNE) of general-sumBayesian games with continuous types and finite actions . Our proof establishes a connection between theequilibria of the infinite Bayesian game and those of finite approximations . This leads to an algorithm to construct a BNE of infinite Bayesiangames by discretizing players’ type spaces .…

Data-driven neural network priors have shown promising performance forRGB natural image denoising . However, data-driven priors are hard to acquire forhyperspectral images (HSIs) due to the lack of training data . A remedy is touse the so-called unsupervised deep image prior (DIP) framework that is based on the classic spatio-spectral decomposition of HSIs .…

This work presents a large-scale analysis of artificial intelligence (AI) and machine learning references within news articles and scientific publications between 2011 and 2019 . Our methods provide new views intopublic perceptions and subject-area expert discussions of AI/ML and greatly exceed the explanative power of prior work .…

This paper investigates the motion planning of autonomous dynamical systems . Markov decision processes (MDP) with unknown transitionprobabilities over continuous state and action spaces . The proposedLDGBA-based reward shaping and discounting schemes for the model-freereinforcement learning (RL) only depend on the EP-MDP states and can overcomethe issues of sparse rewards .…

The third version of the Hypertext Transfer Protocol (HTTP) is currently in its final standardization phase by the IETF . Besides better security and increased flexibility, it promises benefits in terms of performance . We run a large-scalemeasurement campaign toward thousands of websites adopting HTTP/3, aiming at understanding to what extent it achieves better performance than HTTP/2 .…

We demonstrate the capability of the Unified Communication X (UCX)framework to compose a GPU-aware communication layer that serves multiple parallel programming models developed out of the Charm++ ecosystem . We observe increasesin bandwidth of up to 9.6x in Charm++, 10x in AMPI, and 10.5X in Charm4py .…

This thesis proposes techniques aiming to make blockchain technologies andsmart contract platforms practical by improving their scalability, latency, andprivacy . This thesis starts by presenting the design and implementation ofChainspace, a distributed ledger that supports user defined smart contracts and executes user-supplied transactions .…

Using the framework of operator or Calder\’on preconditioning, uniformpreconditioners are constructed for elliptic operators discretized withcontinuous finite (or boundary) elements . They are constructed as the composition of an opposite order operator, discretizing on the sameansatz space, and two diagonal scaling operators .…

We develop a resource theory of symmetric distinguishability . We study the resource theory for two different classes of free operations . The optimal rate of converting oneelementary source to another is equal to the ratio of their quantum Chernoff divergences .…

The last years have seen a surge in models predicting the scanpaths offixations made by humans when viewing images . However, the field is lacking aprincipled comparison of those models with respect to their predictive power . We evaluate many existing models of scanpath prediction on the datasetsMIT1003, MIT300, CAT2000 train and CAT200 test, for the first time giving adetailed picture of the current state of the art of human scanpath predictions .…

We propose using Carrier Sensing (CS) for distributed interference management in millimeter-wave cellular networks . We describe important challenges in using traditional CS in this setting and propose enhanced CSprotocols to address these challenges . We evaluate the downlink coverage probability of our shared mmWavenetwork using simulations as well as numerical examples based on our analysis .…

Machine learning (ML) has been used to develop increasingly accurate link quality estimators for wireless networks . We propose a newtree-based link quality classifier that meets high performance and fairlyclassifies the minority class and, at the same time, incurs low training cost .…

Robustness of machine learning models is critical for security related applications . We propose using characteristics of the empirical datadistribution, both on correlations between the features and the importance of the features themselves . The key idea of our proposed approach is to enable non-uniformperturbations that can adequately represent these feature dependencies during training .…

With vast mmWave spectrum and narrow beam antenna technology, preciseposition location is now possible in 5G and future mobile communications systems . In this article, we describe how centimeterlevel localization accuracy can be achieved, particularly through the use of map-based techniques .…

We consider the scattering of light in participating media composed ofsparsely and randomly distributed discrete particles . The appearance shows distinctgraininess as opposed to the smooth appearance of continuous media . We propose a practical Monte Carlo rendering solution to solve the transfer of energy in discrete participating media .…

Contingency Model Predictive Control (CMPC) anticipates emergency and keeps the controlled system in asafe state that is selectively robust to the identified hazard . This article presents a linear formulation for CMPC, illustrates its keyfeatures on a toy problem, and then demonstrates its efficacy experimentally on a full-size automated road vehicle that encounters a realistic pop-outobstacle .…

The field of dynamical systems is being transformed by the mathematical tools emerging from modern computing and data science . Koopman spectral theory has emerged as a dominant perspective over the past decade . This linear representation of nonlinear dynamics has tremendous potential to enable the prediction,estimation, and control of non linear systems with standard textbook methods developed for linear systems .…

In this work, we are concerned with neural network guided goal-oriented aposteriori error estimation and adaptivity using the dual weighted residualmethod . The primal problem is solved using classical Galerkin finite elements . The proposed algorithm is based on the generalgoal-oriented error estimation theorem .…

The proposed DOTA dataset contains 1,793,658 object instances of 18 categories of oriented-bounding-boxannotations collected from 11,268 aerial images . Previous challenges run on DOTA have attracted more than 1300 teams worldwide . We believe that the expanded large-scale DOTA datasets, the extensive baselines, the code library and the challenges can facilitate thedesigns of robust algorithms and reproducible research on the problem of object detection in aerial images.…

Knowledge distillation (KD) has witnessed its powerful ability in learningcompact models in deep learning field, but it is still limited in distillinglocalization information for object detection . Existing KD methods for objectdetection mainly focus on mimicking deep features between teacher model and student model .…

We propose a novel CP-Net based composition approach to qualitatively select an optimal set of consumers for an IaaS provider . The provider’s and consumers’ qualitative preferences are captured using CP-Nets . A greedy-based and a heuristic-based consumer selection approaches are proposed that effectively reduce the search space of candidates in the composition .…

We introduce Density sketches (DS) as a succinct online summary of the datadistribution . DS can accurately estimate point wise probability density . DS also provides a capability to sample unseen novel data from the underlying data distribution . DS construction is an online algorithm.…

This work introduces a concept of explanations with respect to the violation of safe behaviours within software defined networks . In our setting, a safe behaviour is characterised by aNetKAT policy, or program, which does not enable forwarding packets from aningress i to an undesirable egress e .…

Due to the multiscale nature of our modelequation, the model is challenging to solve by using conventional methods . In our framework, the solution is approximated by a neural network that satisfies both the governingequation and other constraints . The network is then trained with a combinationof different loss terms .…

The proposed MTMS-TCN method outperforms in both phase and step recognition by 1-2% in accuracy, precision and recall, compared to single-task methods . The proposed method shows that the joint modeling of phases and steps is beneficial to improve the overall recognition of each type of activity .…

Deep learning is effective in graph analysis . It is widely applied in many areas, such as link prediction, node classification, communitydetection, and graph classification etc . Graphfool can achieve an average improvement of 11.44% in attack success rate . To the best of our knowledge, this is the first targeted label attack technique .…

Aims to provide absolute lower bounds for the best possible runningtimes that can be achieved by $(1+\lambda)$-type search heuristics on common benchmark problems . We apply our hybrid Monte Carlo dynamic programming approach to aconcatenated jump function and demonstrate how the obtained bounds can be used to gain a deeper understanding into parameter control schemes .…

Constrained reinforcement learning involves multiple rewards that must accumulate to given thresholds . In this class of problems, we show a simple example in which the desired optimal policy cannot be induced by any combination of rewards . This work addresses this shortcoming by augmenting the state with Lagrange multipliers and reinterpreting primal-dualmethods as the portion of the dynamics that drives the multipliers evolution .…

Fault injection technique has been used to test NanosatC-BR-2, a 2-U Cubesat based nanosatellite underdevelopment by INPE . It is very useful technique to test systems prototypes . The FEM is modelled to work at the communication bus emulating eventual faults of the communicatingsubsystems in the messages exchanged .…

Spline interpolation has been used in several applications due to its smoothness and accuracy of the interpolant . When there exists a discontinuity or a steep gradient in the data, artifacts can appear due to the Gibbs phenomenon . Preservation of data monotonicity is a requirement in some applications, and that property is not automatically verified by the interpolator .…

In this paper, we propose a new Sound Event Classification (SEC) method which is inspired in recent works for out-of-distribution detection . In our method,we analyse all the activations of a generic CNN in order to produce featurerepresentations using Gram Matrices .…

In a first-of-its-kind study, this paper proposes the formulation ofconstructing prediction intervals (PIs) in a time series as a bi-objectiveoptimization problem and solves it with the help of Nondominated SortingGenetic Algorithm (NSGA-II) The proposed models when applied to the macroeconomic time series, yielded better results interms of both prediction interval coverage probability (PICP) and predictioninterval average width (PIAW) Compared to the state-of the art Lower UpperBound Estimation Method (LUBE) with Gradient Descent (GD) The 3-stage modelyielded better PICP compared to the 2-stage model but showed similarperformance in PIAW with added computation cost of running NSGA-I second time .…

The continuous speech separation (CSS) is a task to separate the speech sources from a long, partially overlapped recording, which involves a varying number of speakers . A straightforward extension of conventional utterance-level speech separation to the CSS task is to segment the long recording with asize-fixed window and process each window separately .…

Different kinds of models are used to study various natural and technicalphenomena . The paper describes several model approaches which we used in the study of the random early detection algorithm for active queue management . Both the model approaches themselves and their implementation and the results obtained are described .…

If machine failures can be detected preemptively, maintenance and repairs can be performed more efficiently, reducing production costs . Hierarchical Temporal Memory (HTM) outperforms state-of-the-art algorithms at preemptively detecting real-world cases of bearing failures and simulated 3D printer failures . Using the Numenta AnomalyBenchmark, our approach achieves anaverage score of 64.71 .…

We propose a generalized Eulerian-Lagrangian (GEL) discontinuous Galerkin(DG) method . The GEL DG method is motivated for solving linear hyperbolicsystems with variable coefficients . The velocity field for adjointproblems of the test functions is frozen to constant . Numerical results on 1D and 2D linear transport problems are presented to demonstrate great properties of the method .…

An implicit finite difference method isemployed to discretize the Riesz space fractional diffusion equations with apenalty term in a rectangular region by the volume-penalization approach . The spectrum of the preconditioned matrix is also investigated . Numerical experiments are carried out todemonstrate the efficiency of the proposed method, and the stability and convergence of the method are studied .…

The TravelingSalesperson Problem (TSP) and Quadratic Assignment Problem (QAP) are two of the most prominent permutation problems . Theoretical results show most mutation operators for both problems ensure production of maximally diverse populations of sufficiently small size withincubic expected run-time .…

Number of wildfires and area of landburned around the world has been steadily increasing, partly due to climaticchanges and global warming . There is a high probability that more people will be exposed to and endangered by forest fires . EscapeWildFire is an amobile application connected to a backend system which models and predict geographical progression, assisting citizens to escape wildfires inreal-time .…

We propose a new family of high-order explicit generalized-$\alpha$ methods for hyperbolic problems . The user can control the numerical dissipation in the discretespectrum’s high-frequency regions by adjusting the method’s coefficients . We exploit efficient preconditioners for the isogeometric matrix to minimize the computational cost .…

Joint localization algorithm significantly outperforms baselines using only ToA or AoA measurements . It achieves 2-D sub-meter accuracy at the90%-ile. We also numerically demonstrate that the algorithmis more robust to synchronization errors than the baseline using ToAmeasurements only. We evaluate the algorithm performance using aproprietary prototype deployed in an indoor factory environment.…

New hope is placed in Artificial Intelligence (AI) to serve as anunbiased anti-corruption agent . Ever more available (open) government datapaired with unprecedented performance of such algorithms render AI the nextfrontier in anti-corruption . Summarizing existing efforts to use AI-basedanti-corruption tools (AI-ACT), we introduce a conceptual framework to advanceresearch and policy .…

Recently, the end-to-end approach has been successfully applied tomulti-speaker speech separation and recognition in both single-channel and multichannel conditions . However, severe performance degradation is stillobserved in the reverberant and noisy scenarios, and there is still a large performance gap between anechoic and reverberant conditions .…

This work studies the optimization of a wireless sensor network (WNS)at higher dimensions by focusing to the power allocation of decentralizeddetection . To the best of the authors knowledge, this is the first time that LGSO algorithms are applied to the optimal power allocation in IoT networks .…

Self-attention models have been successfully applied in end-to-end speechrecognition systems . However, such attention-based models cannot be used in online speech recognition . Memory-Self-Attention (MSA) Transducer only needs localtime features as inputs . MSA efficiently models long temporal contexts by attending memory states .…