VALENTIN LEPLAT
Assistant Professor in Applied Mathematics
Innopolis University
420500 Innopolis, Russia, 1, Universitetskaya Str.
RESEARCH INTERESTS
Low-rank approximations: nonnegative matrix and tensor factorization
Convex and Non-Convex Numerical Optimization
Numerical linear algebra
EDUCATION AND PROFESSIONAL EXPERIENCE
Nov. 2021 - June. 2023: Senior Research Scientist at Skoltech under the supervision of Profs. Ivan Oseledets and Anh Huy Phan
Mar. 2021 - Nov. 2021: Postdoctoral Research Associate at ICTEAM UCLouvain under the supervision of Prof. Yurii Nesterov
2018-2020: PhD in Applied Mathematics (FPM's, University of Mons)
2015-2017: MSc in Computer Sciences engineering and Applied mathematics (IG, University of Mons)
2012-2018: Aerospace engineer (AEROSTRUCTURE DESIGN, SONACA)
2010-2012: MSc in Mechanical engineering (Gramme Institute)
UPDATES
Starting from the first of July 2024, I move to Innopolis University for a new position of Assistant Professor in the Maths department ! I ll teach during the fall semester the courses of "High-dimensional data analysis" and "Optimisation".
Our recent work with Prof. Clemence Prevost (University of Lille, France) called "Minimum-Volume Non-Negative Block-Term Decomposition: Blind Data Fusion and Unmixing with Estimation of the Number of Endmembers." just got accepted for IEEE EUSIPCO 2024 conference that takes place in Lyon In Augustus 2024.
Our paper entitled "Deep Nonnegative Matrix Factorization with Beta Divergences" just got accpeted for publication in Neural Computation !
With various collaborators from Skoltech, we just released a new peprint which introduces a novel algorithm to compute the Canonical Polyadic Decomposition of an input tensor. The main ingredient of this method relies on the update of two factors of the decomposition at a time, improving the speed of convergence and the overall robustness of the decompsition process compared to state-of-the-art methods in challenging scenerios such as the ones involving highly colinear factor matrices. Check the "Publications tab" :).
The SIAM LA program is now out ! My collaborator Jeremy Cohen (CNRS, France) and I will present our recent works on efficient algorithms for tensor decompositions during this session, you are very welcomed to join us !
In collaboration with Le Thi Khanh Hien and Prof. Nicolas Gillis (UMons, Belgium), we are very excited to anounce we just released a new optimization framework for solving a class of multi-convex optimization problems based on Majorization-Minimization with Extrapolation. Check the "Code" tab for getting access to the code and the "Publications" tab for the link to the arXiv version of the preprint.
In collaboration with Salman Ahmadi (Skoltech), Prof. Anh Huy Phan (Skoltech) and Prof. A. Cichocki, we just released a new preprint introducing a novel collaborative neurodynamic model for computing constrained Tensor Decompositions, with an emphasis on the Canonical Polyadic Decomposition (CPD).
With my collaborator Clemence Prevost (University of Lille, France), we just released a new preprint "Data fusion and unmixing with the regularized nonnegative block-term decomposition: joint problems, blind approach and automatic model order selection. " In this work, we present a new framework that introduces regularized Nonnegative Block-Term decomposition problems and efficient algorithms to ultimately tackle two famous problems in remoting sensing at once, that are the Super Resolution Problem and the Hyprespectral Unmixing. In particular, we consider minimum-volume regularizations that allow to deal with two practical and important challenges; the rank-deficient scenario and the automatic estimation of the number of endmembers when limited prior information is available.
In collaboration with Salman Ahmadi (Skoltech), Prof. Anh Huy Phan (Skoltech) and Prof. A. Cichocki, we just released a new Tensor Network, dubbed as Tubal Tensor Train. We show that this new higher-order tensor decomposition model can be used efficiently for various tasks such as tensor compression and tensor completion.
With my collaborator Jeremy Cohen, we will present our recent works on effective computation of Tensor Decomposition models (preprint released on arXiv) at SIAM LA - 2024 Minisymposium on Tensor Decompositions And Their Applications that will take place in Paris in May 2024.
Great news, our preprint "Direct Exoplanet Detection using L-1 Norm Low-rank Approximation" has been accepted at BeNeLearn 2023 conference !
We released a new version for our arXiv preprint presenting our new stochastic optimizer called "NAG-GS", check out the "Publications" tab to get the arXiv link.
In collaboration with Nicolas Gillis (UMons), Hien Le (UMons) and Akwum Onwunta (University of Lehigh), we just released a new framework aimed at computing deep nonnegative matrix factorizations with or without minimum-volume regularizations. Check out "Code" and "Publications" tabs for more information !
Congratulations to Matthieu Boucquey who successfully passed his master thesis defense on the topic of exact NMF with conic-based formulation along with the use of meta-heuristics to speed the computations. This works was supervised by Professor Francois Glineur from UCLouvain and me. It extends a recent publication on similar topic. Check the tab "Code" for more details.
Good news; our joint work with Clemence Prevost on the topics of the super-resolution problem and hyperspectral unmixing has been accepted at GRETSI 2023 conference.
In collaboration with Prof. Anh Huy Phan and Dr. Salman Ahmadi Asl, we organize a special issue dedicated to Tensor Decompositions and Tensor Networks. You are welcomed to submit your work; all the details are given [here] !
Great news; our paper on higher-order proximal methods for computing nonnegative CPD has been accepted for presentation at IEEE SSP 2023 conference !
I have the priviledge to give a talk at the next data fusion conference (14th April), I'll speak about some recent successes of using higher-order optimization methods for machine learning [abstract]
With collaborators from UCLouvain and UMons (BE), we submitted a new preprint to EUSIPCO 2023 conference on the topic of exoplanet detection by direct imaging.
We submitted a new preprint to IEEE SSP 2023 conference on higher-order proximal methods for computing tensor factorizations.
Good news again, our preprint "Nonnegative Block Term Decomposition with the beta-divergence for data fusion and blind spectral unmixing" (in its short format) has been accepted by ICASSP 2023 conference !
Good news ! Our paper "Conic-Optimization based Algorithms for Nonnegative Matrix Factorization" has been accepted for publication in the journal Optimization Methods and Softwares :).
I ll participate to SIAM Conference on Optimization (OP23) to speak about a new optimization framework for computing nonnegative Tucker decomposition with sparsity constraints (joint work with Jeremy Cohen).
We released a new framework to tackle two fundamental problems in remote sensing which are the hyperspectral unmixing and the HSI-MSI fusion problems. A short version format has been submitted to ICASSP 23, a long version is available in HAL (see publication tab).
We released a new paper where we propose a new stochastic optimizer for large-scale problems. The algorithm is dubbed as NAG-GS
Our Paper "Nonnegative Tucker Decomposition with Beta-divergence for Music Structure Analysis of audio signals" has been presented at Gresti Conference 2022 in Nancy !
Our paper presented at EUSIPCO2021 is now available: 10.23919/EUSIPCO54536.2021.9616335
Today (6th December), our paper "Multi-Resolution Beta-Divergence NMF for Blind Spectral Unmixing " has been accepted by Signal Processing journal.
I will give a talk at the virtual IEEE Asilomar conference 2021 (31th October-3rd Nov.), see the link here
Today (8th October) I move to the Skolkovo Institute of Science and Technology as Senior Research Scientist and postdoc to work with Prof. Ivan Oseledets, Prof. Phan and Prof. Andrzej Cichocki.
Our paper "Fast algorithm for complex-NMF with application to source separation" has been accepted by EUSIPCO 2021 (August 23-27), see you there.
Today (16th March) I move to the Institute of Information and Communication Technologies, Electronics and Applied Mathematics (ICTM), Louvain School of Engineering a postdoc, and work with Prof. Yurii Nesterov
Great news in a short period of time; our paper "Multiplicative Updates for NMF with β-Divergences under Disjoint Equality Constraints" has been accepted by SIMAX for publication !
Our paper on the topic of distributionally robust and multi-objective NMF has been accepted as a Regular Paper in an upcoming issue of the Transactions on Pattern Analysis and Machine Intelligence !
I will give a talk at the virtual SIAM LA21 conference in 2021, see the link here
I successfully passed my PhD public defense on January 18th 2021. The slides and the pdf of the PhD thesis can be downloaded from the page "Publications"- section "THESIS". I thank all the people who attended to the presentation, see you soon !
I passed my PhD private defense on December 8th 2020. My PhD public defense will take place on January 18th 2021; see the link here
Congratulations to my colleague Andersen Man Shun Ang who defended his PhD thesis on October 14; see the video here.
Congratulations to my supervisor Nicolas Gillis for publishing his book on NMF; see the link here