Salem Lahlou

Assistant Professor of Machine Learning

Salem Lahlou | Assistant Professor of Machine Learning

News

  • May 20267 Papers accepted at ICML 2026:
    • Zero-Shot Off-Policy Learning
    • Bayesian Rain Field Reconstruction using Commercial Microwave Links and Diffusion Model Priors
    • Path-dependent Discrete Amortized Inference (Spotlight)
    • Avoid What You Know: Divergent Trajectory Balance for GFlowNets
    • Noise-Corrected GRPO: From Noisy Rewards to Unbiased Gradients
    • CoRe: Collaborative Reasoning via Cross Teaching
    • Position: Agentic AI Systems Should Be Making Bayes-Consistent Decisions
  • April 2026 — Awarded an MBZUAI research grant (PI) for advancing GFlowNet research - Co-PI: Nils Lukas.
  • April 2026 — Accepted invitation to serve as Program Committee Co-Chair for the Frontier AI Systems Workshop (FAISys), Xiamen, China, December 2026.
  • April 2026 — Accepted invitation to serve as Member, Host Scientific Committee for the International Olympiad in Artificial Intelligence (IOAI) 2026, and developed a task for the Main Contest.
  • April 20262 Papers accepted at ACL 2026:
    • FinChain: A Symbolic Benchmark for Verifiable Chain-of-Thought Financial Reasoning
    • SAHM: Arabic Financial Instruction-Tuning Dataset and Models
  • April 2026Paper accepted at Findings of ACL 2026:
    • Temporal Contrastive Decoding: A Training-Free Method for Large Audio-Language Models
  • April 2026Paper accepted at NETYS 2026:
    • Byzantine Cheap Talk: Adversarial Resilience and Topology Effects in LLM Coordination Games
  • March 2026 — Invited Area Chair for NeurIPS 2026.
  • February 2026 — Organized the MBZUAI ML Winter School 2026.
  • January 2026 — Invited talks at the MenaML Winter School 2026 (KAUST) and the AI for Science Symposium (HBKU, Doha).
  • January 2026Paper accepted at EACL 2026 (Oral):
    • Communication Enables Cooperation in LLM Agents: A Comparison with Curriculum-Based Approaches
  • January 2026Paper accepted at Findings of EACL 2026:
    • SD-E2: Semantic Exploration for Reasoning Under Token Budgets
  • December 2025 — Received the Best Presentation Award at ICoIAS 2025 and the Top Reviewer Award at NeurIPS 2025.
  • December 2025Paper accepted at AAAI 2026:
    • Loss-guided Auxiliary Agents for Overcoming Mode Collapse in GFlowNets
  • November 2025 — Invited talk at HSE University's HDI Lab (online).
  • October 2025 — Co-organized the "Bayesianism in the Age of Modern AI" workshop at MBZUAI.
  • October 2025Paper accepted at JMLR 2026:
    • TorchGFN: A PyTorch GFlowNet Library
  • September 2025 — Invited talk at Sorbonne University Abu Dhabi.
  • September 2025 — Invited Area Chair for ICML 2026.
  • September 2025Paper accepted at NeurIPS 2025:
    • SVRPBench: A Realistic Benchmark for Stochastic Vehicle Routing Problem
  • June 2025 — Invited Area Chair for ICLR 2026.
  • May 2025 — Tutorial "Reinforcement Learning through a probabilistic lens" (slides) at the METIS Spring School.
  • May 2025 — Co-organized the International Symposium on Trustworthy Foundation Models at MBZUAI; co-Program Chair at NETYS 2025 (Rabat).
  • April 2025Paper accepted at UAI 2025:
    • On the Privacy Risks of Spiking Neural Networks: A Membership Inference Analysis
  • February 2025 — Co-PI on the EnergyAI flagship research project (2M AED/year, 18 months) and on the Etihad flagship research project (400,000 USD, 12 months).
  • February 2025 — Talks at the MBZUAI–Berkeley Joint Workshop and the MBZUAI Paris Workshop 2025.
  • February 2025Paper accepted at ICLR 2025 Workshop on Bidirectional Human-AI Alignment (to appear in PMLR):
    • Mitigating Societal Cognitive Overload in the Age of AI: Challenges and Directions
  • January 2025Paper accepted at NAACL 2025:
    • PORT: Preference Optimization on Reasoning Traces
  • December 2024 — Keynote at the MoroccoAI 2024 conference.
  • November 2024 — Attended the CLIMB and Domain Adaptation workshops at the Simons Institute, UC Berkeley.
  • September 2024 — Joined MBZUAI as Assistant Professor in the Machine Learning department.