Parallel Algorithm for the Law-of-the-Jungle Learning to the Fast Design of Optimal Codebooks

K. Sano, S. Momose, H. Takizawa, T. Nakajima, C.D. Lima, H. Kobayashi, and T. Nakamura (Japan)

Keywords

parallel processing, vector quantization,optimal codebook design, lawofthejungle algorithm

Abstract

Vector quantization(VQ) is an attractive technique for lossy data compression, which has been a key technology for data storage and/or transfer. So far, various algorithms have been proposed to design optimal codebooks presenting quanti zation with minimized errors. In particular, the Law-of the-Jungle(LOJ) learning algorithm has been proposed to achieve rapid codebook design by applying algorithmic im provements to conventional competitive learning methods. However, the acceleration of the LOJ algorithm is still re quired when large data sets are processed on a single com puter. Accordingly, a scalable codebook design on parallel computers is required. This paper presents a parallel algo rithm for the LOJ learning, suitable for distributed-memory parallel computers with a message-passing mechanism. Ex perimental results indicate a high scalability of the proposed parallel algorithm on the IBM SP2 parallel computer with 32 processing elements.

Important Links:



Go Back