Research

Reinforcement Learning

• <b>Multi-Agent Reinforcement Learning </b> models the interaction of multiple agents as a stochastic game. We proposed communication-based methods to coordinate the behaviors of decentralized agents.

• <b>Offline Reinforcement Learning </b> aims to close the gap between real world environment and simulation environment. We are implementing algorithms on real autonomous vehicles for navigation, sensing, etc.

Social Computing

• <b>Information Diffusion Prediction </b> is to predict the information cascade with historical data. We adopted cascade structure as the main feature for prediction.

AI4Science

• <b>Single cell ATAC clustering </b> is to cluster the single cell ATAC data, which is high-dimension, sparse with lots of dropouts. We employ AI methods to analyze the cells.
• <b>Molecule structure prediction </b> and mining. We use deep (reinforcement) learning to predict the molecule structures.