I
Irina Rish
Researcher at Université de Montréal
Publications - 221
Citations - 7792
Irina Rish is an academic researcher from Université de Montréal. The author has contributed to research in topics: Computer science & Approximation algorithm. The author has an hindex of 34, co-authored 198 publications receiving 6830 citations. Previous affiliations of Irina Rish include IBM & University of California, Irvine.
Papers
More filters
BookDOI
Practical Sparse Modeling An Overview and Two Examples from Genetics
TL;DR: This chapter contains sections titled Sparse Modeling Road Map, Example 1: Genome-Wide Association Studies (GWAS), and Example 2: Gene Microarray Data Analysis.
Proceedings ArticleDOI
A Remedy For Distributional Shifts Through Expected Domain Translation
TL;DR: This work employs multi-modal translation networks to tackle the correlation shifts that appear when data is sampled out-of-distribution and shows that by training a predictor solely on the generated samples, the spurious correlations in training domains average out, and the invariant features corresponding to true correlations emerge.
Posted Content
Scaling Laws for the Few-Shot Adaptation of Pre-trained Image Classifiers
TL;DR: In this article, the authors investigate how the amount of pre-training data affects the few-shot generalization performance of standard image classifiers and find that such performance improvements are well-approximated by power laws as the training set size increases.
Posted Content
Parametric Scattering Networks.
Shanel Gauthier,Benjamin Thérien,Laurent Alsène-Racicot,Irina Rish,Eugene Belilovsky,Michael Eickenberg,Guy Wolf +6 more
TL;DR: In this paper, the scales, orientations, and slants of the wavelet filters are adapted to produce problem-specific parametrizations of the scattering transform, which yield significant performance gains over the standard scattering transform in small sample classification settings.
Journal ArticleDOI
Maximum State Entropy Exploration using Predecessor and Successor Representations
TL;DR: In this paper , the authors propose a method to learn an exploration policy that maximizes the entropy of the state visitation distribution of a single trajectory by conditioning on past episodic experience to make the next exploratory move.