Research Area

Foundation Models & LLMs

Foundation Models & LLMs

• Alignment learning (instruction fine-tuning, RLHF, Imitation learning)

• Complex reasoning (math, code), self-correction, self-reflection

• LLM decoding

• AI-Agents

• Plug-in modules (e.g., RAG)

Machine Learning & Representation Learning

• Uncertatinty measure (measuring confidence of the model output)

• Learning strategies (e.g., self-supervised learning)

• Representation extraction & analysis

• Training model in a restricted environment (Black-box model training & optimization)

• Parameter-efficient training (e.g., pruning, knowledge distillation, quantization)

AI Application

  Explainable AI (XAI)

• SAE-based concept localization in LLMs

• Discovery of CAVs in foundation models

• Visualization of weight propagation for interpretability

RL-based Optimization

• RL for Smart City optimization (HVAC, industrial systems)

• Symbolic & neuro-symbolic policies for interpretable agents

• Rapidly adaptive policies in dynamic environments

• World Models for sim-to-real transfer

• LLM post-training with multi-turn RL & MARL (Multi-Agent RL)

AI for Science

• Material AI: Molecular generation w/ target properties; digital twins via DFT & MD; physics-informed AI for property prediction 

• Climate AI: Extreme Event Tracking & Prediction; AI-driven Nowcasting; Climate Digital Twins; Climate Intervention Modeling (MCB)