Salem Lahlou

Salem Lahlou is an Assistant Professor in the Machine Learning Department at the Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), having joined in September 2024. He previously served as a Senior Researcher at the Technology Innovation Institute (TII) in 2024. His academic background includes a PhD from Mila and Université de Montréal (UdeM) under Yoshua Bengio (2023), with prior studies in applied mathematics at École Polytechnique and statistical learning at École Normale Supérieure Paris-Saclay.

His research focuses on developing more capable and reliable AI systems through three interconnected pillars:

• Novel Method Development: Core contributions to Generative Flow Networks (GFlowNets), uncertainty estimation techniques (DEUP), and curriculum learning frameworks
• Large Language Model Advancement: Enhancing reasoning capabilities and alignment through preference optimization and trace-based learning
• Community Tooling: Creation of torchgfn library for GFlowNets and benchmarks including BabyAI, FinChain, and LLM-BabyBench

Core research areas span Machine Learning, GFlowNets, Uncertainty Estimation, LLM Reasoning, Reinforcement Learning, and AI for Science.

Recent publications (2023-2025) demonstrate strong emphasis on GFlowNet theory/improvements (8+ papers), LLM reasoning evaluation (FinChain, LLM-BabyBench), uncertainty quantification, and societal AI impacts. Key application domains include mathematical reasoning, financial systems, privacy preservation, and cognitive science.

He currently advises graduate students including Junyi (privacy risks in SNNs) and Abhijith (LLM reasoning). His group collaborates with MBZUAI faculty (Nils Lukas, Alham Fikri, Mingming Gong, Martin Takac) and industry partners on projects involving Conversational AI, Personalization, and Affective AI.