Chaotic analog associative memory

This paper proposes chaotic analog associative memory (CAAM), which can handle one-to-many learning pairs composed of analog patterns. Most past associative memory models have considered a binary pattern as the pattern to be learned, which makes it difficult to handle analog patterns. It is also a problem that the superposed pattern is recalled due to interaction between the memorized patterns. Thus, it becomes difficult to handle the association of one-to-many learning pairs, in which multiple patterns can be recalled from a single given pattern. In contrast, the proposed model uses a multiwinner self-organizing neural network (MWSONN), which can handle analog patterns, and realizes association of the analog patterns. In the proposed CAAM, the chaotic neuron is introduced as a part of the network, and one-to-many association is realized by utilizing the dynamic recall power of the chaotic neuron. Computer experiments verify that the association of one-to-many learning pairs composed of analog patterns can be realized, and that the proposed model has high noise immunity and robustness to faults.