Title :
Node compression in simulated annealing networks: an investigation with a scene analysis benchmark
Author :
Tambouratzis, Tatiana
Author_Institution :
Inst. of Inf. & Telecommun., NRCPS Demokritos, Athens
fDate :
2/17/1994 12:00:00 AM
Abstract :
Harmony theory (HT) networks constitute simulated annealing networks which greatly benefit from sparse coding: the precision and cooling procedure of the network improve: however, the increase in size renders each network update slow and computationally expensive. The author examines node compression. Comparisons with a sparse HT network are made, using as a benchmark a scene analysis task
Keywords :
image coding; image recognition; neural nets; simulated annealing; complex constraint propagation problems; cooling procedure; harmony theory networks; node compression; scene analysis benchmark; simulated annealing networks; sparse coding;
Journal_Title :
Electronics Letters
DOI :
10.1049/el:19940214