Convergence of Stochastic Descent Algorithms with Markovian Inputs
Project funded by the Labex CIMI - P.I.: Gersende Fort
Topic: Convergence of Stochastic Descent-based algorithms with Markovian inputs for non-smooth and convex optimization :
Convergence of perturbed iterative algorithms, for convex composite optimization
Acceleration methods
Design and Convergence of Penalized Stochastic Approximation EM algorithms
Iterative algorithms for Quantile and Superquantile estimations.
When: from January 2018 to December 2019
Local participants:
Nicolas Dobigeon (IRIT)
Charles Dossal (IMT)
Cédric Févotte (IRIT)
Gersende Fort (IMT)
Sébastien Gadat (TSE-R)
Aurélien Garivier (IMT puis ENS Lyon)
Laurent Risser (IMT)
External favouref partners:
Jean-François Aujol (IMB)
Emmanuel Gobet (CMAP)
Anatoli Juditski (LJK)
Eric Moulines (CMAP)
Adeline Samson (LJK)
Scientific production:
Papers
G. Fort, E. Ollier and A. Leclerc-Samson. Stochastic Proximal Gradient Algorithms for Penalized Mixed Models. Statistics and Computing, 29(2):231-253, 2019.
G. Fort, L. Risser, Y. Atchadé and E. Moulines. Stochastic FISTA algorithms: so fast ? Proceedings of the IEEE Statistical Signal Processing workshop, pp.796-800, June 2018.
S. Crepey, G. Fort, E. Gobet and U. Stazhynski. Quantification d’incertitude pour l’Approximation Stochastique. Accepté, Actes de conférence du GRETSI 2019.
S. Crepey, G. Fort, E. Gobet and U. Stazhynski. Uncertainty quantification for Stochastic Approximation limits using Chaos Expansion. Accepted for publication in SIAM-ASA Journal on Uncertainty Quantification, 2020.
C. Dossal – xxx
C. Dossal – xxx
L. Risser – xxx
Talks
By G. Fort. Conference IEEE Statistical Signal Processing; Freiburg, Germany; June 2018.
By C. Dossal. 1st Workshop « Advances in nonsmooth analysis and optimization », Erice, Italy; June 2019.
By G. Fort. Conférence francophone GRETSI, Lille; Août 2019.
By G. Fort. PGMO days, December 2019.
By G. Fort. Conference « Optimization for Machine Learning », CIRM, Mars 2020.
Workshop
Co-organization of the semester « Optimization » funded by the Labex CIMI : a first workshop in June 2018 on « Aspects fondamentaux et Exploitation de la structure » with C. Dossal and S. Gadat as co-organizers ; and a second workshop on « Optimization and Learning » co-organized by G. Fort, S. Gadat, A. Garivier, L. Risser and C. Févotte.
Stage
M2 ( see C. Dossal)
And after ?
These CIMI fundings were a cornerstone for its members, in order to apply to other projects funded by national grants:
C. Dossal, G. Fort and S. Gadat are part of a project funded by the french National Research Agency (ANR), starting in Autumn 2019. The P.I. is S. Gadat. It involves researchers from Bordeaux (IMB) and Toulouse (IMT, TSE-R).
G. Fort, S. Gadat and L. Risser are part of a project funded by the Fondation Simone and Cino Del Duca, starting in January 2020. The P.I. is G. Fort
S. Dobigeon, C. Févotte, and L. Risser are members of a national program for Artificial Intelligence, « Artificial and Natural Intelligence Toulouse Institute (ANITI) », starting in Autumn 2019.