Bio

Mouhssine Rifaki is a Graduate Visiting Researcher in Electrical Engineering at Stanford University, working in the Arbabian Lab with Amin Arbabian on algorithmic architectures for adaptive sensing and inference in physical AI systems. He holds an MSc from ENS Paris-Saclay (Mathematics, Vision, Learning – MVA) and a BSc in Mathematics from Sorbonne University. His research aims to show that exploiting latent structure in Markov decision processes can reduce the sample complexity of reinforcement learning by orders of magnitude, enabling agents that learn from minimal data.

Professional Education

  • MSc, ENS Paris-Saclay, Statistics and Machine Learning - MVA (2026)
  • BSc, Sorbonne University, Mathematics (2023)

Contact

  • Academic
    mrifaki@stanford.edu
    University - Student Department: Electrical Engineering Position: Graduate

Additional Info

Links

Current Research and Scholarly Interests

My research aims to show that exploiting latent structure in Markov decision processes can reduce the sample complexity of reinforcement learning by orders of magnitude, enabling agents that learn from minimal data.

I focus on low-rank matrix completion to recover Q-functions from sparse samples via leveraged CUR decompositions; two-to-infinity subspace recovery for the row-wise error control required in policy extraction; score matching in learned latent spaces for diffusion world models with planning guarantees that scale with intrinsic dimension; kernel spectral methods for continuous block MDP decoding; and reducing the demonstration complexity of imitation learning when the expert’s Q-function has low-rank structure.

Projects

  • Adaptive Sensing for Physical AI, Stanford University

    Fast–slow hierarchical architectures, where prediction error from a lightweight model drives modality and resolution selection for a high-fidelity model. Proof of concept on multimodal gesture recognition. Joint work with Amin Arbabian.

    Location

    Stanford, CA

  • Adaptive Population Play for Autonomous Driving, New York University

    Multi-agent reinforcement learning for autonomous driving with co-player conditioning and in-context adaptation to diverse driver populations. Joint work with Eugene Vinitsky (NYU).

    Location

    New York, NY