I am a graduate student at Stanford University in the Institute for Computational and Mathematical Engineering (ICME), formerly in the Applied Mathematics department at Ecole polytechnique, Paris. Prior to entering Ecole polytechnique, I was a Mathematics/CS preparatory class (Classe preparatoire) student at Lycee Sainte-Genevieve in Versailles.
My primary research interest is artificial intelligence. I am especially interested in building systems that can learn tasks from scratch (reinforcement learning) and/or use accumulated knowledge to efficiently learn new tasks (meta-learning). I am currently a research assistant in the Stanford Artificial Intelligence Lab, working with Prof. Emma Brunskill at the intersection of on- and off-policy reinforcement learning methods.
I am also highly interested in the statistical aspects of financial markets. At Ecole polytechnique, I worked with Prof. Jean-Philippe Bouchaud on some applications of control theory to trading. In 2018, I spent 6 months at Tower Research Capital in New York, where I worked on some applications of nonlinear machine learning techniques to high-frequency trading. This year (2019), I spent 3 months at Squarepoint Capital in Singapore, where I worked on several predictive signals on Asian equity markets.
I really enjoy teaching. I was a teaching assistant in Mathematics at Lycee Sainte-Genevieve in the academic year 2016-2017, conducting weekly oral examinations for undergraduates (linear algebra, real analysis, probability and group theory among other things). More recently, I have been a teaching assistant at Stanford for CS221 (AI: Principles and Techniques), in Fall 2018 and Spring 2019.
In my free time, I play a lot of music (acoustic and electric guitar). I also enjoy traveling and solving interesting puzzles.
Publications and preprints
- M. Emschwiller, B. Petit, J.-P. Bouchaud, “Optimal Multi-Asset Trading with Linear Costs: A Mean-Field Approach”, arxiv: 1905.04821 (preprint, 2019).
- B. Petit, L. Amdahl-Culleton, Y. Liu, J. Smith, P.-L. Bacon, “All-Action Policy Gradient Methods: A Numerical Integration Approach”, NeurIPS 2019 Optimization Foundations of Reinforcement Learning Workshop, arXiv: 1910.09093 (2019).
Selected research projects
- Image completion with neural processes (at Stanford, see details here).
- Deep learning under massive label noise (at Tower Research).
- Stochastic trust-region optimization algorithms, applications to traffic models (in collaboration with Aimsun).
- Fair share problems (under the supervision of Prof. De Seguins-Pazzis).
- An essay on the plactic monoid (RSK correspondence and the Erdos-Szekeres theorem).
benpetit [at] stanford [dot] edu