Lisa_seminaires Octobre 2018

lisa_seminaires@iro.umontreal.ca

5 participants
3 discussions

[Tea Talk] Gauthier Gidel (Mila) Fri October 26 2018 11:00 PCM Z315
by Rim Assouel 26 Oct '18

26 Oct '18

This week we have our own Gauthier Gidel giving a talk on Fri October 26 2018 at 11:00 in room PCM Z315 Will this talk be streamed <https://mila.bluejeans.com/809027115/webrtc>? Yes Recorded? Yes you GAN learn a lot going to this talk (too easy?) See you there! Rim and Sai TITLE A Variational Inequality Perspective on Generative Adversarial Networks KEYWORDS GANs, variational inequality, mini-max optimization ABSTRACT Generative adversarial networks (GANs) form a generative modeling approach known for producing appealing samples, but they are notably difficult to train. One common way to tackle this issue has been to propose new formulations of the GAN objective. Yet, surprisingly few studies have looked at optimization methods designed for this adversarial training. In this work, we cast GAN optimization problems in the general variational inequality framework. Tapping into the mathematical programming literature, we counter some common misconceptions about the difficulties of saddle point optimization and propose to extend methods designed for variational inequalities to the training of GANs. We apply averaging, extrapolation and a novel computationally cheaper variant that we call extrapolation from the past to the stochastic gradient method (SGD) and Adam. BIO Gauthier Gidel received the Diplôme de l’École Normale Supérieure in 2017 (ULM MPI2013) and the Master of Science MVA from École Normale supérieur Paris-Saclay in 2016. Gauthier is currently pursuing his PhD at Mila and DIRO from Université de Montréal under the supervision of Simon Lacoste-Julien.Gauthier’s PhD thesis topic revolves around saddle point optimization (a.k.a mini-max problems) for machine learning and more generally variational inequalities on which Gauthier has published several papers [Gidel et al. 2017, Gidel et al. 2018].

2 3

[Tea Talk] Nick Pawlowski (ICL) Fri October 19 2018 10:30 JC S1-111
by Michael Noukhovitch 19 Oct '18

19 Oct '18

This week we have *Nick Pawlowski * from * ICL (interning at FAIR) * giving a talk on *Fri October 19 2018* at *10:30* in *Jean Coutu S1-111* Will this talk be streamed <https://mila.bluejeans.com/809027115/webrtc>? Yes Nick has to be back at FAIR at 2pm so I'd suggest if you want to meet with him then come out to lunch and you can arrange to meet Nick after. *Prior* to seeing this ad, I would have been very *uncertain* about coming. But now it seem like just what the doctor ordered! Michael *TITLE* Bayesian Deep Learning and Applications to Medical Imaging *KEYWORDS *bayesian deep learning, medical applications *ABSTRACT* Deep learning revolutionised the way we approach computer vision and medical image analysis. Regardless of improved accuracy scores and other metrics, deep learning methods tend to be overconfident on unseen data or even when predicting the wrong label. Bayesian deep learning offers a framework to alleviate some of these concerns by modelling the uncertainty over the weights generating those predictions. This talk will review some previous achievements of the field and introduce Bayes by Hypernet (BbH). BbH uses neural networks to parametrise the variational approximation of the distribution of the parameters. We present more complex parameter distribution, better robustness to adversarial examples, and improved uncertainties. Lastly, we present the use of Bayesian NNs for outlier detection in the medical imaging domain, particularly the application of Brain lesion detection. *BIO* Nick is a PhD student in the Biomedical Image Analysis group at Imperial College London, supervised by Ben Glocker. He works on methods to integrate and use uncertainty with deep learning methods. He focuses on Bayesian neural networks and their use for outlier detection. Nick is currently a Research Intern at FAIR Montreal and a main developer of DLTK, a toolkit for deep learning for medical imaging. During this summer he was a Machine Learning resident at Google X.

2 4

[Tea Talk] Devon Hjelm (MSR Montreal + Mila) Fri October 12 2018 10:30 JC S1-111
by Michael Noukhovitch 12 Oct '18

12 Oct '18

This week we have our very own *Devon Hjelm * from * MSR Montreal x Mila * giving a talk on *Friday October 12 2018* at *10:30* in room *Jean Coutu S1-111* Will this talk be streamed <https://mila.bluejeans.com/809027115/webrtc>? No Will this talk be recorded? Yes Maximize your own information and come learn at this talk! Michael *TITLE* Learning representations with Deep InfoMax *KEYWORDS *representation learning, unsupervised learning, adversarial learning *ABSTRACT* In this work, we perform unsupervised learning of representations by maximizing mutual information between an input and the output of a deep neural network encoder. Importantly, we show that structure matters: incorporating knowledge about locality of the input to the objective can greatly influence a representation’s suitability for downstream tasks. We further control characteristics of the representation by matching to a prior distribution adversarially. Our method, which we call Deep InfoMax (DIM), outperforms a number of popular unsupervised learning methods and competes with fully-supervised learning on several classification tasks. DIM opens new avenues for unsupervised learning of representations and is an important step towards flexible formulations of representation-learning objectives for specific end-goals (https://arxiv.org/abs/1808.06670) *BIO* Devon Hjelm is a researcher at Microsoft Research Montreal and an Adjunct Professor at MILA. He did his postdoc at MILA where he focused on adversarial learning and generative models. His current research focuses on using mutual information estimation objectives in representation learning for applications in computer vision, natural language, and RL.

2 2

2024

2023

2022

2021

2020

2019

2018

2017

2016

2015

2014

2013

2012

2011

2010

2009

2008

2007

Lisa_seminaires Octobre 2018