Anton Chechetka
I am a 6th year graduate student at the Robotics
Institute advised by Carlos Guestrin. My research
interests are in machine learning and probabilistic inference.
For the first 2.5 years of the PhD program I was advised by Katia Sycara and worked on algorithms for distributed constraint optimization.
Research
I am interested in principled ways to construct probabilistic models that accurately represent reality and at the same time are feasible for exact inference. More specifically, I am working on learning thin junction trees from data.
Publications:
- Focused Belief Propagation for Query-Specific Inference.
[pdf]
[bib/abs]
Anton Chechetka and Carlos Guestrin.
Artificial Intelligence and Statistics (AISTATS-2010), May 2010.
- Learning Thin Junction Trees via Graph Cuts.
[pdf]
[bib/abs]
Dafna Shahaf, Anton Chechetka and Carlos Guestrin.
Artificial Intelligence and Statistics (AISTATS-2009), April 2009.
- Efficient Principled Learning of Thin Junction Trees.
[pdf]
[bib/abs]
[pdf with proofs]
Anton Chechetka and Carlos Guestrin.
Advances in Neural Information Processing Systems (NIPS 2007), December 2007.
Distributed constraint optimization is a way to formalize the problem of coordination in the group of cooperative agents (for example, robots, people, or sensor nodes). Each agent has exclusive control over one variable and the group has to jointly select an assignment for these variables (e.g. come up with a joint meetings schedule) so as to maximize performance (sum of the values of constraints). Each constraint is a real-valued function over a subset of variables. The agents communicate with each other in order to agree on the jointly optimal assignment.
A popular basic method for solving such problems is multiagent search. To speed up the solution process, we exploit the distributed aspect of the problem by having different agents explore non-intersecting regions of the search space simultaneously. This technique reduces synchronization overhead and makes pruning of the search space faster.
Publications:
- No-Commitment Branch and Bound Search for Distributed Constraint Optimization.
[pdf
]
[bib]
Anton Chechetka and Katia Sycara.
AAMAS-2006 poster session, May 2006.
- An Any-space Algorithm for Distributed Constraint Optimization.
[pdf]
[bib]
Anton Chechetka and Katia Sycara.
AAAI Spring Symposium on Distributed Plan and Schedule Management, March 2006.
- A Decentralized Variable Ordering Method for Distributed Constraint Optimization.
[pdf]
[bib]
Anton Chechetka and Katia Sycara.
AAMAS-2005 poster session, July 2005.
- A Decentralized Variable Ordering Method for Distributed Constraint Optimization.
[pdf]
[bib]
Anton Chechetka and Katia Sycara.
CMU Robotics Institute Technical Report CMU-RI-TR-05-18, May
2005. (detailed version of the AAMAS-2005 paper).
Teaching
In Spring'06 I TA'd 10-701/15-781 Machine Learning class.
Coursework
Here is the list of classes I have taken at CMU so far.