Yiwei Bai
Ph.D. Student
CS Department@Cornell University
Office: 344, Gates Hall, Cornell, Ithaca, NY 14853
Email: yb263 [at] cornell (dot) edu
GitHub /
LinkedIn


Education
Cornell University, USA
Ph.D. in Computer Science, Aug. 2018 to Present


Shanghai Jiao Tong University, China
Bachelor of Engineering, Sep. 2014 to Jun. 2018


Cornell University, USA
Research Intern, June. 2017 to Dec. 2017




Zero Training Overhead Portfolios for Learning to Solve Combinatorial Problems
Yiwei Bai, Wenting Zhao, Carla P. Gomes
Under review.
We have observed that
welltrained models for combinatorial problems acquired in the same training trajectory, with similar top validation performance, perform well on very different validation
instances
ZTop leverages these diverse models to increase the test performance with (almost) zero training overhead.


Batch Learning from Bandit Feedback through Bias Corrected Reward Imputation
Lequn Wang, Yiwei Bai, Arjun Bhalla, Thorsten Joachims
Appears in Realworld Sequential Decision Making workshop, ICML, 2019.
We introduce a new "Model the World" style batch learning from logged bandit feedback algorithm: Bias Corrected Reward Imputation (BCRI).
BCRI learn a rewardregression model and derive a policy from the estimated rewards.
The problem is formulated as bilevel optimization, where the upper level maximizes the DM estimate and the lower lever fits a weighted rewardregression.


Deep Reasoning Networks for Unsupervised Pattern Demixing with Constraint Reasoning
Di Chen, Yiwei Bai, Wenting Zhao, Sebastian Ament, John M. Gregoire, Carla P. Gomes
ICML 2020.
We proposed Deep Reasoning Networks (DRNets), an endtoend framework that combines deep learning with reasoning for solving complex tasks.
DRNets exploit problem structure and prior knowledge by tightly combining logic and constraint reasoning with stochasticgradientbased neural network optimization.
We illustrate the power of DRNets on demixing overlapping handwritten Sudokus and on a substantially more complex task in scientific discovery: CrystalStructurePhaseMapping


Scalable Relaxations of Sparse Packing Constraints: Optimal Biocontrol in PredatorPrey Networks
Johan Bjorcks, Yiwei Bai, Yexiang Xue, Xiaojian Wu, Mark Whitemore, Carla P. Gomes
In Proceedings of the ThirtySecond AAAI Conference on Artificial Intelligence (AAAI), 2018.
In this research, we solved an important problem in computational sustainability  biological control of invasive species.
We proposed an approximation algorithm based on a width relaxation and randomized projections, which is quite scalable compared with previous work and can be used for realistic problem
We evaluated our algorithm in the context of biocontrol for the insect pest Hemlock Wolly Adelgid(HWA) in eastern North America.


An Empirical Study of Collective Behaviors in Manyagent Reinforcement Learning (Extended Abstract)
Yiwei Bai^{*}, Lantao Yu^{*}, Yaodong Yang^{*}, Jun Wang, Weinan Zhang, Ying Wen, Yong Yu
In the Proceedings of the 17th International Conference on Autonomous Agents and Multiagent Systems (AAMAS), 2018.
In this research, we designed and developed Millionlevel MultiAgent Reinforcement Learning Platform. You could find the platform here.
We tried to understand AI population with multiagent reinforcement learning and verify the principles developed in
the real world could be applied to AI population.


Yi, an AI platform playing the GO(game)
Yiwei Bai, Lequn Chen, and colleagues in Tianrang, (advised by Professor Guirong Xue), Jan. 2017
Yi won the Fourth prize in the first International Computer Go Competition
Yi won the Ninth place in the 10th UEC Cup
I and colleagues trained and tuned the policy network and value network
I and colleagues designed and implemented the reinforcement learning framework of the value network incorporated with policy network

