I am a Ph.D. candidate at the Department of Computer Science of Cornell University. My advisor is Prof. Bart Selman.
I got my B.S. degree
in computer science at Peking University in Beijing, China.
Research interestsMy current research focuses on applying stochastic local search methods to probabilistic reasoning. I am also working on
integration of maching learning and reasoning techniques, and developing fast automated reasoning systems.
- A New Approach to Model Counting
Wei Wei and Bart Selman
Eighth International Conference on Theory and Applications of Satisfiability Testing, St. Andrews, Scotland, 2005
We introduce ApproxCount, an algorithm that approximate the number of satisfying assignment of a Boolean formula. Many AI tasks, such as reasoning in Bayesian belief networks, can be reduced to this problem. Our algorithm produces estimates of model counts of formulas much larger than those can be handled by existing algorithms.
- Towards Efficient Sampling: Exploiting Random Walk Strategies
Wei Wei, Jordan Erenrich, and Bart Selman
The Nineteenth National Conference on Artificial Intelligence, San Jose, CA, 2004
We show that random walk style procedures provide a promising technique for sampling from the set of satisfying assignments. Moreover, we demonstrate by combining random walk and Monte Carlo Markov Chain methods how one can get approximately uniform sampling in a number of domains.
- Learn to Speed Up Search
Bart Selman and Wei Wei
AAAI 2002 Workshop on Probabilistic Approaches in Search, Edmonton, Alberta, Canada, 2002
We survey several promising approaches of integrating machine learning techniques into search algorithms. The general methodology of these approaches is to use learning techniques to uncover the hidden structure of the search space, and then use this information to speed up search.
- Accelerating Random Walks
Wei Wei and Bart Selman
Eighth International Conference on Principles and Practice of Constraint Programming, Ithaca, NY, 2002
The algorithm proposed in this paper identifies long distance dependencies among variables in the underlying problem instance. By adding new problem constraints, we made these dependencies explicit, and literally speed up random walk style search in the structured domains.
- Learn to Speed Up Search, AAAI 2002 Workshop on Probabilistic Approaches in Search, Edmonton, Alberta, Canada, 2002
- Accelerating Random Walks, Eighth International Conference on Principles and Practice of Constraint Programming, Ithaca, New York, USA, 2002
- Sampling Combinatorial Space Using Biased Random Walks, Eighteenth International Joint Conference on Artificial Intelligence, Acapulco, Mexico, 2003
- Learn to Walk Faster, AI seminar, Cornell University, 2003
- Towards Efficient Sampling: Exploiting Random Walk Strategies, The Nineteenth National Conference on Artificial Intelligence, San Jose, California, USA, 2004
- Sampling, Counting, and probabilistic inference, AI seminar, Cornell University, 2005
ApproxCount estimates the number of satisfying assignment of a Boolean formula. The algorithm is based on near-uniform sampling of the solution
space of the Boolean formula.
SampleSat samples the solution space of a Boolean formula near uniformly. The code is coming soon.
Contact InfoDepartment of Computer Science
4142 Upson Hall
Ithaca, NY 14853
Email : email@example.com
Last modified: 4/19/05