Systems Seminar

A Factor-Graph Approach to Learning and Data Compression

Dr. Pascal O. Vontobel
ECE Dept.

Abstract

Factor graphs and the sum-product algorithm are tools that can be used to understand various old and new algorithms. One of the most prominent applications is their use in formulating the iterative message-passing decoding of low-density parity-check and turbo codes, two of the best-performing classes of channel codes. Many other algorithms from communications, signal processing, and artificial intelligence can also be cast in the factor-graph/sum-product-algorithm framework.

The main part my talk will be to show that a straight-forward application of the sum-product update rules to a certain factor graph yields one of the best universal lossless data compression algorithms, namely the context-tree weighting (CTW) method, for which Willems, Shtarkov, and Tjalkens received the 1996 IEEE Information Theory Society Best Paper Award. Actually, we will embed this main part into a more general discussion about factor graphs in the context of learning (as e.g. in adaptive filtering and support vector machines).

(No prior knowledge about universal data compression is assumed. Basics from factor graphs and the sum-product algorithm will be briefly reviewed.)

Time and Place: Wed., May 18, at 3:30 pm in 4610 EH

SYSTEMS SEMINAR WEB PAGE: http://homepages.cae.wisc.edu/~gubner/seminar/

File "vontobel2.shtml" last modified Tue 15 Oct 2019, 01:45 PM, CDT
Web Page Contact: John (dot) Gubner (at) wisc (dot) edu