AIDRC Seminar Series - Ivan Oseledets

Title:

Tensor decompositions: algorithms and applications

Abstract:

In this talk, I will present a brief review and cover some of the highlights on the "tensor research" starting from the basics of the tensor-train decomposition and more recent results which involve applications in machine learning and modelling.

Tensors are efficient tools to represent multivariate functions, and many algorithms reduce to classical linear algebra tasks such as low-rank approximation. We will focus mainly on tensor-train decomposition. Important algorithms involve TT-SVD, TT-Cross, Rieamannian optimization. Important applications include solution of high-dimensional PDEs (such as Fokker-Planck), approximation of multivariate functions, compression of neural networks, approximation of multivariate densities and many others.

Bio:

Ivan Oseledets graduated from MIPT in 2006 and defended his PhD in 2007 and Doctor of Sciences (Habilitation) in 2012 both from Institute of Numerical Mathematics of Russian Academy of Sciences. Starting from 2013 he works in Skoltech (Associate Professor, since 2019 - Full Professor, since 2022 - Director of Center for Artificial Intelligence Technologies). His research interests include numerical linear algebra, tensor methods, machine learning, deep learning and artificial intelligence. He is the author of more than 180 papers in reputable international journals and rank A/A* conferences. He is the Associate Editor of SIAM Journal on Scientific Computing, Advances in Computational Mathematics, Inverse and Ill-posed problems. He has been area chair for NeurIPS, ICML and ICLR multiple times.

He is the recipient of Russian President award for Young Scientists (2018), Ilya Segalovich award (2019), SIAM Outstanding paper prize (2018).

Ivan has been a supervisor of 12 PHD candidates who have successfully defended.