moteur de recherche

Séminaire A3SI - Sixin Zhang
Séminaire A3SI - Sixin Zhang
18-Mar-2021 08:30
Age: 34 days

Title: Representation learning for image analysis and modeling Speaker: Sixin Zhang

When & where: March 18th, 8:30AM,

Abstract: Recent breakthrough in image recognition is based on a core idea of representation learning using convolutional neural networks. Remarkably, representations in these networks can also be used to model stationary fields with complex geometric structures, such as textures. I shall present my contributions to answer the following three inter-related questions: What are structures of data? Can certain structures of data be learnt? How to learn?

I shall first review my PhD work on distributed and stochastic algorithms for deep learning over large dataset.
This non-convex optimization problem motivated me to study simplified representations for deep learning models.  I shall present a phase harmonic representation which connects the notion of phase in signal processing with convolutional neural networks. This representation allows to define a multi-scale model to accurately synthesis stationary turbulent flows using wavelets. Lastly, I shall present my recent work on learning low-rank representations with applications to source separation.  

Bio : Sixin Zhang obtained his PhD in Computer Science (2010-2016) at Courant Institute of Mathematical Sciences, NYU, advised by Yann LeCun. After graduation, he worked as a postdoc researcher at ENS Paris in France with Stéphane Mallat on wavelet analysis and deep learning. Then he worked one-year at Peking University (Center for Data Science) in China as a research associate. He is now at CNRS,IRIT,Université de Toulouse as a postdoc researcher in the group of Cédric Févotte.

<- Back to: Accueil