Neural computer uses optical fingers
David R. Selviah · Physics World · 1994
Neural network computers have hardware architectures more like the human brain than those of current digital computers. This makes them much faster at pattern recognition, optimization problems and high-speed data retrieval from vast data–bases. Instead of being programmed, they are “trained” by the repeated presentation of data, usually in the form of images. This saves programming costs but the training time can be prohibitive if neural architectures have to be simulated on serial electronic computers. Faster training speeds are possible with electronic neural hardware, but to achieve fast learning and operation with many neurons optoelectronic hardware is needed. Such a system has now been demonstrated by a group from Trinity College Dublin in collaboration with Hitachi.