Author: Arnaud LEGRAND

Journée au vert POLARIS

DATAMOVE/POLARIS picnic

DATAMOVE/POLARIS BBQ 2019

POLARIS Bootcamp (May 2019)

Journée au vert POLARIS

DATAMOVE/POLARIS picnic

DATAMOVE/POLARIS BBQ 2019

POLARIS Bootcamp (May 2019)

Events in February–March 2023

News

Next seminars

Events

Seminar Polaris-tt Learning in finite-horizon MDP with UCB (Romain Cravic)

Meta

Arnaud LEGRAND 2022/05/23 2022/06/08News

We all had a great day at La Gélinotte. Everyone had the chance to discuss a topic of his choice and it was very instructive. We should do this more often! Continue reading

Arnaud LEGRAND 2021/06/22 2021/06/22News

A real BBQ party with family and friends as we usually do would have been great but a simple post-pandemy picnic was already quite nice! Let’s consider this as a rehearsal! 🙂 Continue reading

Arnaud LEGRAND 2019/06/14 2019/06/14News

Super barbecue avec toutes les familles, les nouveaux, les anciens, …

Arnaud LEGRAND 2019/05/24 2019/06/14News

Internal use: the pad where we took notes on the organization. Les photos Long talks:

Arnaud Legrand: Introduction to Bayesian sampling (sources)
Bary Pradelski: Preferences, Utilities and Identity Economics
Nicolas Gast: Bandits and online …
Continue reading

7 Seminar Polaris-tt Learning in finite-horizon MDP with UCB (Romain Cravic) 2:00 pm – 3:00 pm February 7, 2023 Most of you probably know Markov Decisions Processes (MDP). They are very useful to handle situations where an agent interacts with an environnement that may involve randomness. Concretely, at each time the MDP has a current state and the agent chooses an action : This couple state-action induces a (random) reward and a (random) state transition. If the probability distributions for rewards and transitions are known, at least theoretically, designing optimal behaviors for the agent is easy. What about the case where these distributions are unknown at the early stage of the process ? How to LEARN optimal behaviors efficiently ? A popular way to handle this issue is to use the optimism paradigm, inspired from UCB algorithms designed for stochastic bandits problems. In this talk, I will expose the main ideas of two possible approaches, UCRL algorithm and optimistic Q-learning algorithm, that use optimism to well perform in finite-horizon Bâtiment IMAG (406) Saint-Martin-d'Hères, 38400 France Map More information Read more