Title :
Neural-net inference and content addressable memory
Author :
Mazza, Christian
Author_Institution :
Section de Math., Geneva, Switzerland
fDate :
1/1/1997 12:00:00 AM
Abstract :
We consider Whittle´s probabilistic content addressable memory, the antiphon, which is designed for recovering stored patterns from nonlinear distortions of the input messages. We give an application to content-based image retrieval systems and propose canonical ways of choosing similarity thresholds ensuring statistical consistency
Keywords :
content-addressable storage; inference mechanisms; neural nets; probability; antiphon; content-based image retrieval systems; neural-net inference; nonlinear distortions; probabilistic content-addressable memory; similarity thresholds; statistical consistency; Associative memory; Content based retrieval; Hebbian theory; Hopfield neural networks; Hypercubes; Image retrieval; Maximum likelihood decoding; Neural networks; Quantum computing; Testing;
Journal_Title :
Neural Networks, IEEE Transactions on
