| 摘要 |
The connecting matrix between input and output neurons is adjusted to reduce errors between the desired output and the actual output, and the number of adaptive elements are reduced so that large scaled optical implementation is done easily. The optical neural network includes a multi-layer hologram (47) for interconnecting input neurons and output neurons, spatial light modulators (3,8) for providing local adaptive interconnection of neurons, a foward path (1) for adaptive learning, a reverse path (2) for calculating error transmitted reversly, a two dimensional detector (5) for calculating sigmoid function in the output plane, a two-dimensional detector (10), for calculating error in the input stage, and a personal computer (6) for designating strength of local adaptive interconnection of the spatial light modulators (3,8) .
|
