Computational and statistical trade-offs in learning

22 mars 2016, 08:45 → 23 mars 2016, 19:00 Europe/Paris

Salle de Conférences Marilyn et James Simons (Le Bois-Marie)

COMPUTATIONAL AND STATISTICAL TRADE-OFFS IN LEARNING

Organized by: Sylvain Arlot (Université Paris-Sud, Paris-Saclay), Francis Bach (INRIA Paris), Alain Celisse (Université de Lille 1)

This workshop focuses on the computational and statistical trade-offs arising in various domains (optimization, statistical/machine learning).
This is a challenging question since it amounts to optimize the performance under limited computational resources, which is crucial in the large-scale data context.
One main goal is to identify important ideas independently developed in some communities that could benefit the others.

Speakers :

Pierre Alquier (ENSAE, Paris-Saclay)
Alexandre d'Aspremont (D.I., CNRS/ENS Paris)
Quentin Berthet (DPMMS, Cambridge Univ., UK)
Alain Celisse (Université de Lille 1)
Rémi Gribonval (INRIA, Rennes)
Emilie Kaufmann (CNRS, Lille)
Vianney Perchet (CREST, ENSAE Paris-Saclay)
Garvesh Raskutti (Wisconsin Institute for Discovery, Madison, USA)
Ohad Shamir (Weizmann Institute, Rehovot, Israel)
Silvia Villa (Istituto Italiano di Tecnologia, Genova & MIT, Cambridge, USA)

 

Abstracts abstracts.pdf

Poster SCHLUMBERGER.pdf

mardi 22 mars ¶

08:45 → 17:00 mardi 22 mars¶

09:00 Café d'accueil

10:00 Ohad Shamir¶

Trade-offs in Distributed Learning

Transparents shamir.pdf

11:00 Alain Celisse¶

Using kernels to detect abrupt changes in time series

Transparents celisse.pdf

12:00 Déjeuner

13:30 Alexandre d'Aspremont¶

Renegar's Condition Number and Compressed Sensing Performance

Transparents aspremont.pdf

14:30 Emilie Kaufmann¶

Optimal Best Arm Identification with Fixed Confidence

Transparents kaufmann.pdf

15:30 Pause café

16:00 Vianney Perchet¶

Highly-Smooth Zero-th Order Online Optimization

Transparents perchet.pdf

mercredi 23 mars ¶

09:00 → 17:00 mercredi 23¶

09:00 Café d'accueil

10:00 Garvesh Raskutti¶

Algorithmic and statistical perspectives of randomized sketching for ordinary least-squares

Transparents raskutti.pdf

11:00 Pierre Alquier¶

On the Properties of Variational Approximations of Gibbs Posteriors

Transparents alquier.pdf

12:00 Déjeuner

13:30 Quentin Berthet¶

Trade-offs in Statistical Learning

Transparents berthet.pdf

14:30 Remi Gribonval¶

Projections, Learning, and Sparsity for Efficient Data Processing

Transparents gribonval.pdf

15:30 Pause café

16:00 Silvia Villa¶

Generalization properties of multiple passes stochastic gradient method

Transparents villa.pdf