Time+Place: Tuesday 05/04/2016 14:30 Room 337-8 Taub Bld.
Title: Designing Communication Receivers Using Machine Learning Techniques
Speaker: Brian M. Kurkoski - COLLOQUIUM LECTURE http://www.jaist.ac.jp/is/labs/bits/brian
Affiliation: Japan Advanced Institute of Science and Technology (JAIST), Nomi, Japan
Host: Eitan Yaakobi


Any communication receiver, for all its complexity, can be decomposed
into a sequence of operations on an input received value sequence
to obtain an output estimated transmitted sequence. Such receivers
are widely used in modern forms of wireless communications and data
storage. Practical implementations are in VLSI, where continuous or 
floating-point values are quantized to a fixed number of bits.

The objective is to directly design fixed-precision operations to 
maximize mutual information.  For the design of communication systems, 
maximization of mutual information gives the highest possible 
communications rate, and thus is a natural choice as a design criteria.  
Each operation in a sequence of decoding operations can be implemented 
using a lookup table, and the core idea is to design lookup tables that 
maximize mutual information.  We call this the "max-LUT method."

To do this, we leverage some well-known results from machine learning 
theory.  In particular, the design of classifiers that minimize 
conditional entropy is equivalent to the design of quantizers that 
maximize mutual information.  To avoid exponential complexity, a 
variation of the K-means  clustering algorithm, called KL-means 
algorithm, is used to find the quantizer (classifier).  There is a 
natural connection between a quantizer and a lookup table.  These ideas 
are applied to the implementation of fixed-precision algorithms for 
decoding error-correcting codes, particularly low-density parity check 

Short Bio: 
Brian M. Kurkoski is an Associate Professor at the Japan Advanced 
Institute of Science and Technology (JAIST) in Nomi, Japan. Born in 
Portland, Oregon, he received the B.S. degree from the California 
Institute of Technology in 1993, and then worked at two California 
startups. He received the M.S. and Ph.D. degrees from the University of 
California San Diego in 2000 and 2004, respectively. He received a JSPS 
Postdoctoral Fellowship from 2004 to 2006, while at the University of 
Electro-Communications in Tokyo, Japan, where he continued as Associate 
Professor from 2007 to 2012. He has been at JAIST since 2012. He was an 
associate editor for IEICE Transactions on Fundamentals of Electronics, 
Communications and Computer Sciences from 2010 to 2014. He is currently 
the secretary of the Data Storage Technical Committee, a technical 
committee of the IEEE Communications Society.  He has organized or been 
technical co-chair for various conference tracks and workshops on coding 
for data storage. His research interests include coding theory, 
information theory and communication theory and their applications.

Refreshments will be served from 14:15
Lecture starts at 14:30