Bonjour à tous,
Vous êtes invité à assister à la présentation d'examen prédoc de Li
Yao qui aura lieu demain.
Titre: Deep Learning and its Applications
Lieu/Date: mardi 10 septembre, 16h, local 3195 (AA)
Résumé:
I will firstly discuss an application of machine learning in video
games. In particular, a deep multi-layer neural network is trained
to predict the results of a first-person shooting game, compared
with other types of models such as linear regression and trees.
This predictor is used in a match-making system to organize players
into different teams.
In the second part, I will discuss a new probabilistic
interpretation of denoising autoencoders. This new interpretation
uses Markov
chain Monte Carlo sampling to approximate the underlying data
generating distribution and has been proved both theoretically and
experimentally to work well. It is viewed as a special case of a
Generative Stochastic Network.
For the future work, I will discuss some possible extension to the
current Generative Stochastic Networks models. One improvement
could be to use a multimodal reconstruction distribution that would
potentially lead to better samples. I will also discuss briefly how
to use these unsupervised trained models in the classification
tasks.