Top > Seminars & Events > Seminar & Events > Colloquiums (FM & IMI) > Information Theory Based on a Hash Property

Colloquiums (FM & IMI)

Information Theory Based on a Hash Property

Hold Date
2014-07-16 16:45~2014-07-16 17:45
Place
Lecture Room L-1 (3F, IMI, Kyushu University)
Object person
 
Speaker
Jun MURAMATSU (NTT Communication Science Laboratories, NTT Corporation)

The IMI Colloquium in July
 
Date:Wednesday, 16 July 2014
          16:45-17:45 lecture
Place:Lecture Room L-1 (3F, IMI, Kyushu University)
Speaker:Jun MURAMATSU (NTT Communication Science Laboratories, NTT Corporation)
 
Title:Information Theory Based on a Hash Property
 
Abstract:
In 1948, C. E. Shannon established the information theory, which clarifies the limits of the performance (the Shannon limits) for transmission codes such as the data compression codes and the error correcting codes. However, the optimal code constructions are impractical because the time complexity and the memory size are exponential with respect to the dimension n of source sequences and channel inputs. Recently, codes using sparse matrices (LDPC/LDGM codes), which is tractable with large n, are studied actively. However, it is necessary for the optimality of these codes that a source/channel has symmetry. In this study, we can construct optimal codes based on a hash property of an ensemble of functions. In particular, we can prove that code constructions using sparse matrices can achieve the Shannon limits for a general source/channel. In this talk, the outline of the study is explained by using the channel coding problem (error correcting codes) as an example.
 


IMI Colloquium Report