News
December 2010.
Will attend the NIPS conference.
Bio
I'm a PhD student at the University of Montreal.
I'm studying machine learning in the LISA lab,
under the supervision of Yoshua Bengio and
Pascal Vincent. Here's my resume.
I'm interested in sparsity, improving learning in deep architectures and simply finding new applications
to ML.
Code
- Large scale online PCA This python online pca estimator
does online estimation of the leading eigenvectors of the covariance matrix of some samples. Try
it if PCA is too computationally expensive. The trick is to maintain a low rank approximation of
the covariance which is updated periodically by solving a much smaller, but related, eigenvalue
problem (that of the Gram matrix).
- Stacked Denoising Autoencoders We've open-sourced the code I worked on at Google.
It's mainly an implementation of denoising stacked autoencoders for
Torch (a C++ machine learning library) and it's called
deepTorch.
Papers
- Stacked Denoising Autoencoders: learning useful representations in a deep network with a local denoising criterion, Pascal Vincent, Hugo Larochelle, Isabelle Lajoie, Yoshua Bengio and Pierre-Antoine Manzagol, JMLR 2010.
- Why Does Unsupervised Pre-training Help Deep Learning?, Erhan, Bengio, Courville, Manzagol, Vincent, and Bengio, JMLR 2010.
- MIREX Special Tagatune Evaluation Submission, Pierre-Antoine Manzagol and Samy Bengio at the
MIREX 2009 competition.
- The Difficulty of Training Deep Architectures and the Effect of Unsupervised Pre-Training,
Dumitru Erhan, Pierre-Antoine Manzagol, Yoshua Bengio, Samy Bengio, Pascal Vincent at
AISTATS 2009.
- On the Use Of Sparse Time-Relative Auditory Codes For Music, Pierre-Antoine Manzagol,
Thierry Bertin-Mahieux and
Douglas Eck at
ISMIR 2008. Best student paper award!
- Extracting and Composing Robust Features with Denoising Autoencoders,
Pascal Vincent, Hugo Larochelle, Yoshua Bengio, and Pierre-Antoine Manzagol at
ICML 2008.
- Topmoumoute Online Natural Gradient Algorithm,
Nicolas Le Roux, Pierre-Antoine Manzagol
and Yoshua Bengio,
NIPS 2007. Tonga is
a fast online approximation of the natural gradient.
Contact You can contact me at username[at]iro[dot]umontreal[dot]ca, the username being manzagop.