Best Arm Identification: A Unified Approach to Fixed Budget and Fixed Confidence
- Victor Gabillon, M. Ghavamzadeh, A. Lazaric
- Computer ScienceNIPS
- 3 December 2012
A performance bound is proved for the two versions of the UGapE algorithm showing that the two problems are characterized by the same notion of complexity.
Best-Arm Identification in Linear Bandits
- Marta Soare, A. Lazaric, R. Munos
- Computer ScienceNIPS
- 22 September 2014
The importance of exploiting the global linear structure to improve the estimate of the reward of near-optimal arms is shown and the connection to the G-optimality criterion used in optimal experimental design is pointed out.
Linear Thompson Sampling Revisited
- Marc Abeille, A. Lazaric
- Computer ScienceInternational Conference on Artificial…
- 1 November 2016
Thompson sampling can be seen as a generic randomized algorithm where the sampling distribution is designed to have a fixed probability of being optimistic, at the cost of an additional $\sqrt{d}$ regret factor compared to a UCB-like approach.
Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning
- Denis Yarats, R. Fergus, A. Lazaric, Lerrel Pinto
- Computer ScienceInternational Conference on Learning…
- 20 July 2021
DrQ-v2 is able to solve complex humanoid locomotion tasks directly from pixel observations, previously unattained by model-free RL, and provides significantly better computational footprint compared to prior work.
Risk-Aversion in Multi-armed Bandits
- Amir Sani, A. Lazaric, R. Munos
- Computer ScienceNIPS
- 3 December 2012
This paper introduces a novel setting based on the principle of risk-aversion where the objective is to compete against the arm with the best risk-return trade-off, which proves to be more difficult than the standard multi-arm bandit setting.
Learning Near Optimal Policies with Low Inherent Bellman Error
- A. Zanette, A. Lazaric, Mykel J. Kochenderfer, Emma Brunskill
- Computer ScienceInternational Conference on Machine Learning
- 29 February 2020
We study the exploration problem with approximate linear action-value functions in episodic reinforcement learning under the notion of low inherent Bellman error, a condition normally employed to…
Frequentist Regret Bounds for Randomized Least-Squares Value Iteration
- A. Zanette, David Brandfonbrener, Matteo Pirotta, A. Lazaric
- Computer ScienceInternational Conference on Artificial…
- 1 November 2019
It is proved that the frequentist regret of RLSVI is upper-bounded by $\widetilde O(d^2 H^2 \sqrt{T})$ where d are the feature dimension, H is the horizon, and T is the total number of steps.
Finite-sample analysis of least-squares policy iteration
- A. Lazaric, M. Ghavamzadeh, R. Munos
- MathematicsJournal of machine learning research
- 2012
A performance bound is reported for the widely used least-squares policy iteration (LSPI) algorithm based on the performance of the LSTD solution evaluated at the states generated by the Markov chain and used by the algorithm to learn an estimate of the value function.
Online Stochastic Optimization under Correlated Bandit Feedback
- M. G. Azar, A. Lazaric, Emma Brunskill
- Computer ScienceInternational Conference on Machine Learning
- 3 February 2014
The high-confidence tree (HCT) algorithm is introduced, a novel anytime χ-armed bandit algorithm, and regret bounds matching the performance of state-of-the-art algorithms in terms of the dependency on number of steps and the near-optimality dimension are derived.
Reinforcement Learning of POMDPs using Spectral Methods
- K. Azizzadenesheli, A. Lazaric, Anima Anandkumar
- Computer ScienceAnnual Conference Computational Learning Theory
- 25 February 2016
This work proposes a new reinforcement learning algorithm for partially observable Markov decision processes (POMDP) based on spectral decomposition methods and proves an order-optimal regret bound with respect to the optimal memoryless policy and efficient scaling withrespect to the dimensionality of observation and action spaces.
...
...