Commit d37a8445 authored by Loïc Barrault's avatar Loïc Barrault
Browse files

Added GANs

parent 1eeeaa57
%!TEX root = m2_DL_GANS.tex
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%\begin{frame}
% \frametitle{}
%\begin{block}{}
% \begin{itemize}
% \item TEMPLATE
% \end{itemize}
%\end{block}
%\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Modèles discriminants vs. modèles génératifs}
%A discriminative model learns a function that maps the input data (x) to some desired output class label (y). In probabilistic terms, they directly learn the conditional distribution P(y|x).
%A generative model tries to learn the joint probability of the input data and labels simultaneously, i.e. P(x,y). This can be converted to P(y|x) for classification via Bayes rule, but the generative ability could be used for something else as well, such as creating likely new (x, y) samples.
\begin{block}{Modèle discriminant}
\begin{itemize}
\item Apprend une fonction qui fait correspondre les données d'entrée $\vx$ à une certaine classe de sortie $y$
\item modélise la probabilité conditionnelle $p(y|\vx)$
\end{itemize}
\end{block}
\begin{block}{Modèle génératif}
\begin{itemize}
\item Apprend une fonction estimant la distribution de probabilités jointe des entrées $\vx$ et des sorties $y$
\item Modélise la distribution de probabilités jointe $p(\vx, y)$
\item Peut être utilisé pour la classification :
\begin{itemize}
\item Règle de Bayes: $p(\vx, y) = p(y | \vx)p(\vx)$
\end{itemize}
\item[\ra] MAIS l'intérêt réside dans la possibilité de \textbf{créer/générer} de nouveaux exemples (\vx,y)
\item Possibilité de \textbf{comprendre} et \textbf{expliquer} la structure sous-jacente des données d'entrée
\item Sans même avoir de données annotées !
\end{itemize}
\end{block}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{GAN : formalisation}
\begin{block}{}
\begin{itemize}
\item Soit :
\begin{itemize}
\item $q(\vx)$ : distribution marginale des \textbf{données}
\item[\ra] On ne la connait pas ! + on ne veut pas poser d'a priori sur sa forme
\item $p(\vx)$ : la distribution marginale du \textbf{modèle}
\begin{itemize}
\item $p(\vx) = \int p(\vz)p(\vx|\vz) d\vz$
\end{itemize}
\end{itemize}
\item Les propriétés suivantes affectent la taille de sortie $o_j$ d'une couche convolutionnelle selon l'axe $j$:
\end{itemize}
\begin{textblock*}{100mm}[0,0](100mm,30mm)
\only<1>{\includegraphics[valign=t, width=0.5\textwidth]{figures/gan_architecture_1} }
\only<2>{\includegraphics[valign=t, width=0.5\textwidth]{figures/gan_architecture_2} }
\only<3>{\includegraphics[valign=t, width=0.5\textwidth]{figures/gan_architecture_3} }
\only<4>{\includegraphics[valign=t, width=0.5\textwidth]{figures/gan_architecture_full} }
\end{textblock*}
\end{block}
\end{frame}
\documentclass[aspectratio=169,t,xcolor=table]{beamer}
%\documentclass[t]{beamer}
%\documentclass[handout,t]{beamer}
% pdf2ps cm_parole.pdf;
% intro, codage + DTW: psselect -p 1-35,51-61-63 cm_parole.ps > cm_parol_poly.ps
% DTW alone: psselect -p 1-19,35-47 cm_parole.ps > cm_parole_poly.ps
% decode psselect -p1-47,51-74 cm_parole.ps > cm_parole_poly.ps
% psnup -4 -H96mm -W128mm -m15mm -b6mm cm_parole_poly.ps cm_parole_poly.ps4
%
%\usepackage{pgfpages}
%\pgfpagelayout{4 on 1}{a4paper,landscape}
\mode<presentation>
{
%\usetheme{PaloAlto}
% \usetheme{Hannover}
\usetheme{informatics}
\useoutertheme{infolines}
% \setbeamercovered{transparent} % or whatever (possibly just delete it)
}
\def\swidth{.7cm}
\setbeamersize{sidebar width left=\swidth}
\setbeamertemplate{sidebar left}
{
{\usebeamerfont{title in sidebar}%
\vskip1.5em%
\usebeamercolor[fg]{title in sidebar}%
\insertshorttitle[width=\swidth,center,respectlinebreaks]\par%
\vskip1.25em%
}%
{%
\usebeamercolor[fg]{author in sidebar}%
\usebeamerfont{author in sidebar}%
\insertshortauthor[width=\swidth,center,respectlinebreaks]\par%
\vskip1.25em%
}%
\hbox to2cm{\hss\insertlogo\hss}
\vskip1.25em%
\insertverticalnavigation{\swidth}%
\vfill
\hbox to2cm{\hskip0.6cm\usebeamerfont{subsection in
sidebar}\strut\usebeamercolor[fg]{subsection in
sidebar} }
\vskip3pt%
}%
%\insertframenumber/\inserttotalframenumber\hfill}%
\setbeamertemplate{footline}[page number]{} % change the numbering to page numbering instead of frame
\setbeamertemplate{navigation symbols}{}
\usepackage[french]{babel}
\usepackage[utf8]{inputenc}
\usepackage{times}
\usepackage{epsfig}
\usepackage{comment}
\usepackage{url}
\usepackage{multirow}
\usepackage[T1]{fontenc}
%\usepackage{natbib}
%\usepackage{multimedia}
\usepackage{lmodern}
\usepackage{array}
\setlength{\extrarowheight}{3pt}
%\usepackage{xspace}
%\usepackage{amsmath}
\usepackage{graphicx}
\usepackage[export]{adjustbox}
\input ../macros.tex
\usepackage[absolute,showboxes,overlay]{textpos}
%\TPshowboxestrue % commenter une fois fini
\TPshowboxesfalse % décommenter pour faire disparaitre les boites
\textblockorigin{10mm}{10mm} % origine des positions
% This is only inserted into the PDF information catalog. Can be left out.
\subject{Apprentissage Automatique}
\title[]{[Apprentissage Automatique]\\ Generative Adversarial Networks}
\author[]{Loïc Barrault}
\institute[LIUM, Le Mans Université]
{
loic.barrault@univ-lemans.fr \\
Laboratoire d'Informatique de l'Université du Maine \\
}
%\date{09 janvier 2017}
% If you have a file called "university-logo-filename.xxx", where xxx
% is a graphic format that can be processed by latex or pdflatex,
% resp., then you can add a logo as follows:
%\pgfdeclareimage[height=0.5cm]{limsi-logo}{limsilogo}
%\logo{\pgfuseimage{limsi-logo}}
%\logo{\includegraphics[height=0.5cm]{limsilogo}}
%\logo{\epsfbox{limsilogo.eps}}
% Delete this, if you do not want the table of contents to pop up at
% the beginning of each subsection:
%\AtBeginSubsection[]
%{
% \begin{frame}<beamer>
% \frametitle{Outline}
% \tableofcontents[currentsection,currentsubsection]
% \end{frame}
%}
% If you wish to uncover everything in a step-wise fashion, uncomment
% the following command:
%\beamerdefaultoverlayspecification{<+->}
\newtheorem{conclusion}[theorem]{Conclusions}
\begin{document}
\begin{frame}
\titlepage
\end{frame}
%\input{mt_neural.tex}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%\begin{frame}
% \frametitle{}
%\begin{block}{}
% \begin{itemize}
% \item TEMPLATE
% \end{itemize}
%\end{block}
%\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Plan}
\begin{block}{}
\begin{itemize}
\item Modèles discriminants vs. génératifs
\item GAN
\begin{itemize}
\item Architecture
\item Applications
\begin{itemize}
\item Vision
\end{itemize}
\end{itemize}
\end{itemize}
\end{block}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\begin{frame}
\frametitle{Sources principales}
\begin{block}{}
\begin{itemize}
\item "Generative Adversarial Networks", I. Goodfellow\\
\begin{itemize}
\item \url{https://arxiv.org/abs/1406.2661}
\end{itemize}
\item[]
\item "Introduction GAN + code tensorflow",
\begin{itemize}
\item \url{http://blog.aylien.com/introduction-generative-adversarial-networks-code-tensorflow/}
\end{itemize}
\item[]
\end{itemize}
\end{block}
\end{frame}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\input{gan.tex}
% and kill the abominable icon
\setbeamertemplate{bibliography item}{}
\begin{frame}[allowframebreaks]
\frametitle{References}
% \bibliographystyle{amsalpha}
\bibliographystyle{apalike}
\bibliography{refs}
\end{frame}
\end{document}
@article{collobert2011nlp,
author = {Ronan Collobert and
Jason Weston and
L{\'{e}}on Bottou and
Michael Karlen and
Koray Kavukcuoglu and
Pavel P. Kuksa},
title = {Natural Language Processing (almost) from Scratch},
journal = {CoRR},
volume = {abs/1103.0398},
year = {2011},
url = {http://arxiv.org/abs/1103.0398},
archivePrefix = {arXiv},
eprint = {1103.0398},
timestamp = {Wed, 07 Jun 2017 14:43:14 +0200},
biburl = {http://dblp.org/rec/bib/journals/corr/abs-1103-0398},
bibsource = {dblp computer science bibliography, http://dblp.org}
}
@article{kim2014cnnnlp,
author = {Yoon Kim},
title = {Convolutional Neural Networks for Sentence Classification},
journal = {CoRR},
volume = {abs/1408.5882},
year = {2014},
url = {http://arxiv.org/abs/1408.5882},
archivePrefix = {arXiv},
eprint = {1408.5882},
timestamp = {Wed, 07 Jun 2017 14:40:07 +0200},
biburl = {http://dblp.org/rec/bib/journals/corr/Kim14f},
bibsource = {dblp computer science bibliography, http://dblp.org}
}
@article{kalchbrenner2014cnn,
author = {Nal Kalchbrenner and
Edward Grefenstette and
Phil Blunsom},
title = {A Convolutional Neural Network for Modelling Sentences},
journal = {CoRR},
volume = {abs/1404.2188},
year = {2014},
url = {http://arxiv.org/abs/1404.2188},
archivePrefix = {arXiv},
eprint = {1404.2188},
timestamp = {Wed, 07 Jun 2017 14:42:24 +0200},
biburl = {http://dblp.org/rec/bib/journals/corr/KalchbrennerGB14},
bibsource = {dblp computer science bibliography, http://dblp.org}
}
@inproceedings{kalchbrenner2013,
title = "Recurrent Continuous Translation Models",
author = "Kalchbrenner, Nal and Blunsom, Phil",
year = "2013",
address = "Seattle",
journal = "Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing",
month = "October",
publisher = "Association for Computational Linguistics",
}
@inproceedings{dumoulin2016,
title = {A guide to convolution arithmetic for deep learning},
author = {Dumoulin, Vincent and Visin, Francesco},
year = {2016},
}
Generative-Adversarial-Network-Tutorial @ 3307a5bc
Subproject commit 3307a5bc54ef45a73b81490d7aaff90a35bc8d18
Supports Markdown
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment