Yassine Chemingui

I am a PhD candidate at Washington State University, advised by Prof. Jana Doppa.

Research Interests

My work focuses on offline decision-making: learning to optimize and act safely from logged data when new experiments are costly. I build methods that turn static datasets into effective optimization models and control policies, motivated by the need for robust and safe AI in domains like materials discovery, healthcare, and smart grids.

• Offline Safe Reinforcement Learning: Ensuring safety constraints, optimizing robust policies from limited data, and enabling risk-aware decision-making. For fixed budgets,we used a minimax formulation with bandit-style updates; and for variable test-time constraints, we designed adaptive policies that switch at deployment without retraining.
• Offline Model-Based Optimization: Developing surrogate models and efficient search strategies for high-dimensional design problems. This includes recasting offline optimization as an RL problem to guide the search, and designing search-bias-guided surrogates that better align the model's ranking with the optimization goal.

Publications and Preprints