My research focuses on statistical machine-learning methods for systems whose decisions shape the data they subsequently learn from, especially when valid estimation of the target requires accounting for latent variables or auxiliary functions that must themselves be estimated. In these settings, inference and optimization are coupled: the mechanism producing the data determines both which outcomes are observed and what can be learned about alternative interventions. I study how to learn policies, characterize intervention effects—including structured and distributional effects—and quantify uncertainty under this dependence. A central theme in my inference work is nuisance robustness: I design procedures whose leading statistical behavior is insensitive to estimation errors in outcome models, propensities, bridge functions, or learned representations. This enables the use of flexible machine-learning components to improve practical performance without compromising inferential validity. My work combines semiparametric and high-dimensional statistics with causal inference, online learning, kernel methods, and representation learning.

About Me

I am a Research Fellow at the Gatsby Computational Neuroscience Unit at University College London, where I work with Arthur Gretton.

Before joining UCL, I was a postdoctoral researcher in the MIND team at Inria, where I worked on nuisance-robust causal inference and mediation analysis for biomedical applications. I completed my PhD jointly with Inria Thoth and the Criteo AI Lab, focusing on offline policy learning and sequential learning. Earlier, at the Institute for Infocomm Research, I worked on deep generative models, anomaly detection, and medical imaging.

CV · PhD thesis

Feel free to reach out if you wish to collaborate, exchange ideas, or seek Master's thesis supervision. Contact: (first initial).(last name) [at] ucl.ac.uk.

News

• June 2026: Organized the Advances on Adaptive Experimentation (AAE) Workshop at the Gatsby Unit, UCL.
• June 2026: Our paper Fast Best-in-Class Regret for Contextual Bandits was accepted at UAI 2026.

Selected Publications

• Efficient Inference after Directionally Stable Adaptive Experiments
  Z. Shen*, H. Zenati*, N. Kallus, A. Gretton, K. Khamaru, A. Bibaut.
  Preprint, 2026.
  [arXiv]

• Kernel Treatment Effects from Adaptively Collected Data
  Houssam Zenati, Bariscan Bozkurt, Arthur Gretton.
  AISTATS, 2026.
  [Paper] · [Code]

• Semiparametric Efficient Test for Interpretable Distributional Treatment Effects
  Houssam Zenati, Arthur Gretton.
  Preprint, 2026.
  [Paper]

• Fast Best-in-Class Regret for Contextual Bandits
  Girard, Samuel, Nathan Kallus, Jill-Jênn Vie, Arthur Gretton, Aurélien Bibaut, and Houssam Zenati.
  UAI, 2026.
  [arXiv]

• Sequential Counterfactual Risk Minimization
  Houssam Zenati, Eustache Diemert, Matthieu Martin, Julien Mairal, Pierre Gaillard.
  ICML, 2023.
  [Paper] · [Code]

© Copyright 2026 Houssam Zenati. Last updated: July, 2026.