Thursday, January 14, 2016

Bach, Leonardo, Cancer and the Search for the Master Algorithm. A short summary of last night's Paris Machine Learning Meetup (#5 Season 3)

So yesterday we had the Paris Machine Learning Meetup #5 Season 3 and had many interesting presentations:

Amine mentioned the data gathering that is currently going on at The Mathworks to feed data for a webinar he is setting up. More information can be found on the presentation.

Chloe talked to us about Fat Data in Large p, small n: Feature selection with few samples in high dimension. Interestingly she goes beyond a linear regression + sparsity assumption and explores a regularized relevance of sorts. Chloe mentioned that the argument eventually yields a graph-cut optimization. I wonder how complexity arguments and attendant phase transitions play a role in this new formulation ( see ... There will be a "before" and "after" this paper ... )

Sebastien described the work of a few teams that are doing well in the on-going AutoML challenge. Suffice to say Random Forest on Hyperparameters seem to be winning the day among other things. Fascinating. Sebastien's presentation is here.  We covered this challenge a year ago in “Sunday Morning Insight: The Hardest Challenges We Should be Unwilling to Postpone.”
The meetup took a special twist with two other presentations for which results could be directly gauged by a non-specialist audience. Aurore first shared with us her algorithm that learns how to Compose Music with an RBM and an LSTM. Here is the result of what was learned with 67 hours of training music. It was fascinating as one the audience member mentioned that it looked similar to the proficency level his daughter had gotten after 2+ years of piano training.
Finally, Kiran  talked about the current state of the art in Modeling of Human Faces and the hurdles a potential ML system would have to overcome to please Human taste.  Time will tell if a Deep Learning algorithm can reach these human level threshold of quality.Some videos presented at the meetup can be found here.

The video of the meetup is here, except for Kiran's presentation at the very end, all the talks are in French.

All in all, a big thank you to our host and sponsor, the Maltem Consulting Group

This entry is also on Medium and on LinkedIn
Join the CompressiveSensing subreddit or the Google+ Community or the Facebook page and post there !
Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.

No comments: