Mathematics × Computer science
Research
Mathematical foundations of learning and decision-making under uncertainty.
Sequential decisions
How should we gather information and act when decisions have to be made under uncertainty? My work studies dynamic resource allocation, bandit problems and Markov decision processes, including best-arm identification, the exploration–exploitation trade-off, and risk-aware planning and learning.
Recent work concerns distributional reinforcement learning: estimating and optimizing the distribution of returns, beyond their expectation.
Statistical learning
My interests include the mathematical foundations of machine learning, statistical estimation and testing, and the constraints under which learning takes place. Topics include differential privacy, adaptivity, neural networks and high-dimensional probability.
Information theory
Source coding and universal coding, including non-parametric models, countably infinite alphabets and pattern coding. I am also interested in the connections between information-theoretic quantities, statistical inference and sequential decision-making.
Stochastic models & model selection
Context trees and variable-length Markov chains, hidden Markov models, particle methods and perfect simulation. For model selection, my work includes Minimum Description Length, oracle inequalities and the BIC criterion.
Research software
- The exhaustive Peres transform: randomness extraction and universal source coding.
- Best-arm identification in Julia, accompanying work with Emilie Kaufmann.
- Bandit algorithms in Python and MATLAB, with Olivier Cappé and Emilie Kaufmann.
- Baum–Welch implementations: a comparison across programming languages.
- Perfect simulation for processes with long memory.
Research projects
Research funding has included ANR SPADRO (2013–2017), ANR ALICIA (2013–2017), the SeqALO AI chair (2020–2025) and PEPR IA (2023–2027).
