New learning strategies from the microscopic level of an artificial neural network
暂无分享,去创建一个
optical correction signal was applied. These specific features are needed in hardwares realising self-organising neural networks. A computer simulation of XOR learning was performed to confirm the adaptive functions of optical adaptive devices in a circuit based on the back-propagation (BP) algorithm. The circuit consisted of the analogue three-layer network, which was constructed with optical adaptive devices, sigmoid and derivative functions circuits and multipliers. The self-organising processes were computed using analogue functions [6] of the circuits. It was confirmed that the analogue three-layer network including the optical adaptive devices was self-organised without the aid of conventional computers to memorise W, and to calculate correction signals. The self-organising process in the analogue circuit was similar to that calculated with the BP algorithm. Conclusions: An optical adaptive device was developed, whose threshold is varied by an optical correction signal. The synaptic weight W, was varied from -0.7 to +0.7 during operation proportionally to the applied duration of the optical correction signal. The synaptic weight W, was kept in a nonvolatile manner when no optical correction signal was applied. It was confirmed that the self-organisation can be realised using the optical adaptive device in neural networks.
[1] James L. McClelland,et al. James L. McClelland, David Rumelhart and the PDP Research Group, Parallel distributed processing: explorations in the microstructure of cognition . Vol. 1. Foundations . Vol. 2. Psychological and biological models . Cambridge MA: M.I.T. Press, 1987. , 1989, Journal of Child Language.
[2] Geoffrey E. Hinton,et al. Learning internal representations by error propagation , 1986 .