HKUST · IEDA · Interpretable Evaluation Lab
IE Lab — understanding decisions
to improve them
Department of Industrial Engineering and Decision Analytics
We develop rigorous frameworks to evaluate and explain decisions made by algorithms, AI systems, and human policies — grounded in causal inference, mechanistic interpretability, and stochastic optimization. Our work is driven by real-world problems in online platforms, experimentation, and AI deployment.
Causal InferenceIdentifying causal effects from both observational data and randomized experiments — with methods that handle confounding, interference, and distribution shift across platforms and policy settings.
Mechanistic InterpretabilityCircuit-level analysis, sparse autoencoders, and structural causal models to understand how AI systems internally represent knowledge and arrive at decisions.
Policy EvaluationAssessing how well a decision policy performs — from offline data, observational records, or simulation — to inform better policy design in healthcare, operations, and AI systems.
Experimental DesignPrincipled experimental frameworks for online platforms — addressing interference, anomalous data, and valid statistical inference under complex market dynamics.
Reinforcement LearningLearning optimal sequential decision strategies from interaction and feedback — with foundations in reward design, exploration, and generalization across dynamic environments.
Online PlatformsModeling two-sided markets, recommendation systems, and large-scale experimentation infrastructure to inform platform-level policy and resource allocation.
Distributional RobustnessOptimization methods that remain reliable under distribution shift — addressing model misspecification, covariate shift, and out-of-distribution generalization.
SimulationStochastic simulation, physics-based simulation, agent-based simulation, and LLM-driven simulation — to evaluate policies and understand complex system behavior.
