Abstract :
A learning algorithm is produced for the Tank and Hopfield signal decomposition decision circuit of graded-response artificial neurons (see D.W. Tank and J.J. Hopfield, ibid., vol.GAS-33, vol.5, p.533-41, May 1986). By using the algorithm, the circuit can be updated to recognize a new waveform after operation has begun. The proposed algorithm has the following advantages. First, basic waveforms already learned need not be maintained, because these are not needed for learning a new one. Only the new waveform and its basic waveform number are needed. Secondly the learning process is completed in a short time, because it does not include repetitive presentation of basic waveforms. The algorithm is not applicable to other neural networks
Keywords :
learning systems; neural nets; signal processing; circuit update; graded-response artificial neurons; learning algorithm; neural networks; signal decision circuit; signal decomposition; waveform recognition; Circuits and systems; Digital arithmetic; Digital signal processing; Feedback; Frequency response; Microcomputers; Resistors; Signal processing algorithms; Signal resolution; Voltage;
