Memorizing oscillatory patterns in the analog neuron network

An inverse problem relating to associative memory, namely, that of finding a weight matrix such that a network has periodic attractors with the given output waveforms, is investigated. One solution to this problem is given by adaptive neural oscillator (ANO) learning. The ANO is a recurrent network of continuous-time, continuous-output model neurons. Modified back-propagation learning is performed so as to make the output waveform as similar as possible to the external input waveform. If the output waveform sufficiently resembles the input waveform, by using the output feedback waveform instead of that of the external input, the network continues an autonomous oscillation with a waveform similar to the previously given external input one. By combining ANO learning with the scheme of the associative memory network, multiple oscillatory waveforms can be stored in one neural network and can be selectively regenerated with the initial state of the network.<>