Huizhen Janey Yu Department of Computer Science P.O. Box 68 (Gustaf Hällströmin Katu 2b) FIN-00014 University of Helsinki Office: Exactum A240 Tel: (09) 191 51401 Email: janey.yu at cs.helsinki.fiI joined HIIT as a postdoctoral researcher. I graduated from LIDS, Laboratory for Information and Decision Systems, M.I.T. My Ph.D. thesis is on Completely and Partially Observable Markov Decision Processes (MDP and POMDP).
My research interests include: Stochastic control, Machine learning, and Optimization methods.
Teaching:
Spring 2008: Seminar on Predicting Structured Data
Fall 2006: Seminar on Markov Decision Processes
Publication
Ph.D. Thesis: "Approximate Solution Methods for Partially Observable Markov and Semi-Markov Decision Processes," M.I.T., Cambridge, MA, February, 2006. (Abstract , thesis pdf)
D. P. Bertsekas and H. Yu. "A Unifying Polyhedral Approximation Framework for Convex Optimization," LIDS report 2820, M.I.T., September 2009. Submitted. (.pdf)
H. Yu and D. P. Bertsekas. "Basis Function Adaptation Methods for Cost Approximation in MDP," IEEE International Symposium on Adaptive Dynamic Programming and Reinforcement Learning (ADPRL), 2009, Nashville, USA. (.pdf)
H. Yu and D. P. Bertsekas. "New Error Bounds for Approximations from Projected Linear Equations," Technical report C-2008-43, Dept. Computer Science, Univ. of Helsinki, July 2008; revised July 2009. (Also, LIDS report 2797, M.I.T.) Submitted. (.pdf)
-- A shorter/abridged version appeared at European Workshop on Reinforcement Learning (EWRL'08), 2008, Lille, France. (.pdf)H. Yu and J. Rousu. "An Efficient Method for Large Margin Parameter Optimization in Structured Prediction Problems," Technical report C-2007-87, Dept. Computer Science, Univ. of Helsinki, 2007. (.pdf)
D. P. Bertsekas and H. Yu. "Solution of Large Systems of Equations Using Approximate Dynamic Programming Methods," LIDS report 2754, M.I.T., June 2007. (.pdf) Related talk at INFORMS: (.pdf)
-- Journal version: "Projected Equation Methods for Approximate Solution of Large Linear systems." (.pdf) Appeared in J. Computational and Applied Mathematics 227(1), pp. 27-50, May 2009.H. Yu and D. P. Bertsekas. "A Least Squares Q-Learning Algorithm for Optimal Stopping Problems," LIDS report 2731, M.I.T., December 2006. (.pdf)
-- A preliminary version: "Q-learning Algorithms for Optimal Stopping Based on Least Squares," European Control Conference, 2007, Kos, Greece. (.pdf) Presentation slides: (.pdf)H. Yu and D. P. Bertsekas. "Convergence Results for Some Temporal Difference Methods Based on Least Squares," LIDS report 2697, M.I.T., June 2006. (.pdf) Appeared in:
IEEE Trans. Automatic Control 54(7), pp. 1515-1531, July 2009.H. Yu and D. P. Bertsekas. "On Near Optimality of the Set of Finite-State Controllers for Average Cost POMDP," LIDS report 2689, M.I.T., May 2006. Appeared in:
Mathematics of Operations Research 33(1), pp. 1-11, February 2008. (.pdf)H. Yu. "A Function Approximation Approach to Estimation of Policy Gradient for POMDP with Structured Policies," The 21st Conference on Uncertainty in Artificial Intelligence, 2005. (.pdf -- a version with appendices included)
H. Yu and D. P. Bertsekas. "Discretized Approximations for POMDP with Average Cost," The 20th Conference on Uncertainty in Artificial Intelligence, 2004, Banff, Canada. (.pdf)
H. Yu and D. P. Bertsekas. "Lower Approximations of the Optimal Cost Function for Discounted and Average Cost Partially Observable Markov Decision Processes," LIDS Technical report in preparation, M.I.T.
H. Yu and W. E. L. Grimson. "Combining Configurational and Statistical Approaches in Image Retrieval," The 2nd IEEE Pacific-Rim Conference on Multimedia, 2001.