
I have graduated from
the Department of
Computer Science
at the University of
Chicago with a Ph.D.
I defended my thesis "Adaptive Bayesian Inference for Graphical Models" in 2012.
In this work, I developed an adaptive inference method for large data sets, that speeds up
computations from linear to logarithmic time in the size of the problem after a small
modification to the graphical model.
My approach is to leverage the previous computations when an
application repeatedly solves a variation of the same statistical
inference problem. I applied my methods to predict sidechain packing
in proteins from 300 to 5,000 amino acids with an average improvement
of 6.88 times in computation speed. My method is amenable to
parallelism which I applied to improve a widely used approximate
inference algorithm where we decompose the problem into simpler
inference subproblems that are solved repeatedly with minor change.
This work is coauthored by my advisors:
Umut Acar ,
Ramgopal Mettu , and
Alexander T. Ihler .
My departmental advisor was László Babai,

Publications

Ozgur Sumer, Umut A. Acar, Alexander T. Ihler, Ramgopal Mettu
Adaptive Exact Inference in Graphical Models
JMLR Journal of Machine Learning Research Vol. 12, 2011.
Ozgur Sumer, Umut A. Acar, Alexander T. Ihler, Ramgopal Mettu
Fast Parallel and Adaptive Updates for DualDecomposition Solvers
AAAI 25th Conference on Artificial Intelligence (AAAI), 2011.
Ozgur Sumer, Umut A. Acar, Alexander T. Ihler, Ramgopal Mettu
Maintaining MAP Configurations with Applications to Protein Sidechain Packing
IEEE/SP 15th Workshop on Statistical Signal Processing (SSP), 2009.
Ozgur Sumer, Umut A. Acar, Alexander T. Ihler, Ramgopal Mettu
Adaptive Inference on General Graphical Models
Uncertainity in Artificial Intelligence (UAI), 2008.
Ozgur Sumer, Umut A. Acar, Alexander T. Ihler, Ramgopal Mettu
Adaptive Bayesian Inference.
Conference on Neural Information Processing Systems (NIPS), 2007.
Ozgur Sumer
Partial Covering of Hypergraphs.
Symposium on Discrete Algorithms (SODA), 2005.

