Youssef Allouah

I am a postdoctoral researcher at Stanford University. I previously earned my PhD at EPFL after graduating from Ecole Polytechnique. I also interned at Amazon.

My research develops foundations and methods for making machine learning systems trustworthy when data are private, participants are unreliable, or some data must later be forgotten. I study privacy, robustness, and machine unlearning, including in distributed learning systems.

Youssef Allouah

Publications

Distributional Machine Unlearning via Selective Data Removal [paper][code]
Y. Allouah, R. Guerraoui, S. Koyejo.
ICLR 2026; Oral at MUGen@ICML 2025.
What to Forget in Unlearning? Forget Set Curation for Language Models
A. Jha, A. Khatua, Y. Allouah, S. Koyejo.
FoGen@ICML 2026.
The Distillation Game: Adaptive Attacks & Efficient Defenses [paper][code]
Y. Allouah*, M. Haghifam*, S. Koyejo, R. Shokri. (*equal contribution)
FoGen@ICML 2026.
Efficient Prediction of Pass@k Scaling in Large Language Models
J. Kazdan, R. Schaeffer, Y. Allouah, C. Sullivan, K. Yu, N. Levi, S. Koyejo.
arXiv 2025.
Adaptive Gradient Clipping for Robust Federated Learning
Y. Allouah, R. Guerraoui, N. Gupta, A. Jellouli, G. Rizk, J. Stephan.
ICLR 2025, Spotlight.
The Cost of Trust in Machine Learning: Privacy, Robustness, Unlearning, and their Interactions [thesis]
Y. Allouah.
EPFL doctoral thesis, 2025.
The Utility and Complexity of In- and Out-of-Distribution Machine Unlearning [paper][slides]
Y. Allouah, J. Kazdan, R. Guerraoui, S. Koyejo.
ICLR 2025.
Towards Trustworthy Federated Learning with Untrusted Participants [paper]
Y. Allouah, R. Guerraoui, J. Stephan.
ICML 2025.
Certified Unlearning for Neural Networks [paper][code]
A. Koloskova*, Y. Allouah*, A. Jha, R. Guerraoui, S. Koyejo. (*equal contribution)
ICML 2025.
Byzantine-Robust Federated Learning: Impact of Client Subsampling and Local Updates
Y. Allouah, S. Farhadkhani, R. Guerraoui, N. Gupta, R. Pinot, G. Rizk, S. Voitovych.
ICML 2024.
Fine-Tuning Personalization in Federated Learning to Mitigate Adversarial Clients
Y. Allouah, A. El Mrini, R. Guerraoui, N. Gupta, R. Pinot.
NeurIPS 2024.
Revisiting Ensembling in One-Shot Federated Learning
Y. Allouah, A. Dhasade, R. Guerraoui, N. Gupta, A.-M. Kermarrec, R. Pinot, R. Pires, R. Sharma.
NeurIPS 2024.
Robust Sparse Voting [paper][code]
Y. Allouah, R. Guerraoui, L. Hoang, O. Villemaud.
AISTATS 2024.
The Privacy Power of Correlated Noise in Decentralized Learning [paper][code]
Y. Allouah, A. Koloskova, A. El Firdoussi, M. Jaggi, R. Guerraoui.
ICML 2024.
Fixing by Mixing: a Recipe for Optimal Byzantine ML under Heterogeneity [paper]
Y. Allouah, S. Farhadkhani, R. Guerraoui, N. Gupta, R. Pinot, J. Stephan.
AISTATS 2023.
On the Privacy-Robustness-Utility Trilemma in Distributed Learning [paper][video]
Y. Allouah, R. Guerraoui, N. Gupta, R. Pinot, J. Stephan.
ICML 2023.
Robust Distributed Learning: Tight Error Bounds and Breakdown Point under Data Heterogeneity [paper][code][video]
Y. Allouah, R. Guerraoui, N. Gupta, R. Pinot, G. Rizk.
NeurIPS 2023, Spotlight.
Further Results on Latent Discourse Models and Word Embeddings
S. Khalife, D. Goncalves, Y. Allouah, L. Liberti.
JMLR 2021.