Hi Gang,

This Thursday we will have Tomáš Mikolov talk about his recent work with recurrent neural networks. Hope to see you all there.

Time & Date: 14h00 Thursday July 21st.
Location: LISA Lab (AA3256)

Title: Recent advances in neural network based language modeling

Abstract:

I plan to describe novel techniques for training neural network language
models. I will introduce the recurrent architecture, class based neural
net model, and simple maximum entropy model that can be trained jointly
with the neural net model. Together, these techniques allow training of
very good language models on hundreds of millions of words. Results
after application to the state of the art speech recognition systems will
be presented, as well as comparison to other popular language modeling
techniques.

Next, I'd like to present some ideas how to further extend and improve
the existing models, and how to possibly make them closer to the human
performance. I hope for some discussion about this part.


Cheers,
Aaron

--
Aaron C. Courville
Département d’Informatique et
de recherche opérationnelle
Université de Montréal
email:Aaron.Courville@gmail.com