Search

In this paper we propose a novel method that provides contrastiveexplanations justifying the classification of an input by a black boxclassifier such as a deep neural network. Given an input we find what should beminimally and sufficiently present (viz. important object pixels in an image)to justify...

Cartograms are maps that rescale geographic regions (e.g., countries,districts) such that their areas are proportional to quantitative demographicdata (e.g., population size, gross domestic product). Unlike conventional baror pie charts, cartograms can represent correctly which regions share...

We study a natural problem in graph sparsification, the Spanning TreeCongestion (\STC) problem. Informally, the \STC problem seeks a spanning treewith no tree-edge \emph{routing} too many of the original edges. The root ofthis problem dates back to at least 30 years ago, motivated by applications...

A stochastic model predictive controller (SMPC) of air conditioning (AC)system is proposed to improve the energy efficiency of electric vehicles (EV).A Markov-chain based velocity predictor is adopted to provide a sense of thefuture disturbances over the SMPC control horizon. The sensitivity...

Gauge-invariance is a fundamental concept in physics---known to provide themathematical justification for all four fundamental forces. In this paper, weprovide discrete counterparts to the main gauge theoretical concepts, directlyin terms of Cellular Automata. More precisely, we describe a step-by...

We give a maximal independent set (MIS) algorithm that runs in $O(\log \log\Delta)$ rounds in the congested clique model, where $\Delta$ is the maximumdegree of the input graph. This improves upon the $O(\frac{\log(\Delta) \cdot\log \log \Delta}{\sqrt{\log n}} + \log \log \Delta )$ rounds algorithm...

This paper presents an efficient technique to prune deep and/or wideconvolutional neural network models by eliminating redundant features (orfilters). Previous studies have shown that over-sized deep neural networkmodels tend to produce a lot of redundant features that are either shiftedversion of...

A distributed binary hypothesis testing problem is studied in which multiplehelpers transmit their observations to a remote detector over orthogonaldiscrete memoryless channels. The detector uses the received samples from thehelpers along with its own observations to test for the joint distribution...

Deep convolution networks have proved very successful with big datasets suchas the 1000-classes ImageNet. Results show that the error rate increases slowlyas the size of the dataset increases. Experiments presented here may explainwhy these networks are very effective in solving big recognition...

We construct efficient, unconditional non-malleable codes that are secureagainst tampering functions computed by small-depth circuits. Forconstant-depth circuits of polynomial size (i.e. $\mathsf{AC^0}$ tamperingfunctions), our codes have codeword length $n = k^{1+o(1)}$ for a $k$-bitmessage. This...

Many journals post accepted articles online before they are formallypublished in an issue. Early citation impact evidence for these articles couldbe helpful for timely research evaluation and to identify potentially importantarticles that quickly attract many citations. This article investigates...

Long simulation times in climate sciences typically require coarse grids dueto computational constraints. Nonetheless, unresolved subscale informationsignificantly influences the prognostic variables and can not be neglected forreliable long term simulations. This is typically done via...

We address the problem of compactly storing a large number of versions(snapshots) of a collection of keyed documents or records in a distributedenvironment, while efficiently answering a variety of retrieval queries overthose, including retrieving full or partial versions, and evolution historiesfor...

Practical application of H[infinity] robust control relies on systemidentification of a valid model-set, described by a norm-bounded differentialinclusion, which explains all possible behavior for the control plant. This isusually approximated by measuring the plant repeatedly and finding a model...

We present an approximation algorithm that takes a pool of pre-trained modelsas input and produces from it a cascaded model with similar accuracy but loweraverage-case cost. Applied to state-of-the-art ImageNet classification models,this yields up to a 2x reduction in floating point multiplications...

We study the computational complexity of ARRIVAL, a zero-player game on$n$-vertex switch graphs introduced by Dohrau, G\"{a}rtner, Kohler,Matou\v{s}ek, and Welzl. They showed that the problem of deciding terminationof this game is contained in $\text{NP} \cap \text{coNP}$. Karthik C. S.recently...

Medical visualization is the use of computers to create 3D images frommedical imaging data sets, almost all surgery and cancer treatment in thedeveloped world relies on it.Volume visualization techniques includesiso-surface visualization, mesh visualization and point cloud visualizationtechniques...

Semidefinite programming can be considered over any real closed field,including fields of Puiseux series equipped with their nonarchimedeanvaluation. Nonarchimedean semidefinite programs encode parametric families ofclassical semidefinite programs, for sufficiently large values of theparameter...

Patch-based low-rank minimization for image processing attracts muchattention in recent years. The minimization of the matrix rank coupled with theFrobenius norm data fidelity can be solved by the hard thresholding filter withprinciple component analysis (PCA) or singular value decomposition (SVD)...

We explore recurrent encoder multi-decoder neural network architectures forsemi-supervised sequence classification and reconstruction. We find that theuse of multiple reconstruction modules helps models generalize in aclassification task when only a small amount of labeled data is available...