Title :
Rotor Associative Memory with a Periodic Activation Function
Author :
Suzuki, Yozo ; Kitahara, Michimasa ; Kobayashi, Masaki
Author_Institution :
Interdiscipl. Grad. Sch. of Med. & Eng., Univ. of Yamanashi, Kofu, Japan
Abstract :
Complex-valued Associative Memory (CAM) can store multi-state patterns unlike Hopfield Associative Memory (HAM). CAM stores not only given training patterns but also many spurious patterns, such as their rotated patterns, at the same time. Rotor Associative Memory (RAM) can make the most of rotated patterns unstable but the reversed patterns remain stable. In the present work, we propose RAM with a Periodic Activation Function (PAF) to make the reversed patterns unstable. PAF is an activation function that Aizenberg introduced to CAM. We prove that RAM with a PAF has far fewer spurious patterns by using dynamic associative memories which can search the stored patterns.
Keywords :
content-addressable storage; learning (artificial intelligence); pattern recognition; CAM; PAF; RAM; complex-valued associative memory; dynamic associative memories; multistate patterns; periodic activation function; rotor associative memory; spurious patterns; stable reversed patterns; stored pattern search; training patterns; unstable rotated patterns; Associative memory; Computer aided manufacturing; Computer simulation; Neurons; Random access memory; Rotors; Training; associative memory; complex-valued neural networks; periodic activation function; rotor associative memory;
Conference_Titel :
Neural Networks (IJCNN), The 2012 International Joint Conference on
Conference_Location :
Brisbane, QLD
Print_ISBN :
978-1-4673-1488-6
Electronic_ISBN :
2161-4393
DOI :
10.1109/IJCNN.2012.6252462