Pattern Separation by Chaotic Networks Employing Chaotic Itinerancy
Takeshi Yamakawa, Tsutomu Miki, Masayoshi Shimono · Industrial and Engineering Applications of Artificial Intelligence and Expert Systems · 2022
An ordinary neural network can store information in the form of distributed connective weights and thresholds, and facilitates readout of the information in the form of training output data by applying input data similar to but different from the training input data. Although storing behavior in the training mode in the network is dynamical, the readout behavior is static. Accordingly, there cannot be searching of alternative candidates to be read out. This is quite different from information retrieval in a biological brain. The brain retrieves events one by one, by chaining a train of thought or by analogy of different fields.