Projects

Machine Learning for Combinatorial Optimization

NeurIPS 2021 Competition, Jun 2021 ~ Oct 2021
Advisor: Prof. Shipra Agrawal
Collaborator: Abhi Gupta, Sian Lee, Luocen Wang

• Improved the branch-and-bound algorithm by replacing the heuristic branching policy with a sophisticatedly trained graph convolutional neural network model
• Developed a training method in a combination of behavior cloning and parallel evolutionary algorithm
• Designed and built a Singularity container for running and evaluating algorithms on the high-performance computing cluster

Amazon Last Mile Routing Research Challenge

Amazon & MIT Research Challenge, Mar 2021 ~ Jun 2021
Advisor: Prof. Sharon Di
Collaborator: Chen Jing, Yunhao Xing

• Realized an attention model in the Transformer architecture for route optimization problems under realistically sized problem instances
• Trained the model via reinforcement learning and imitation learning; Conducted experiments to compare different training methods
  certificate,