Folthretの電子回路化

抄録

One of the common problems in the neuro-computer hardwares is complication of wiring. Folthret as a basic element for the neuro-computer, which has been offered by one of the authors of this paper, can solve this type of problem quite well. Therefore this element is advantageous to make up the neuro-computer. In this paper, Folthret has been at first realized with the hybrid style of analog and digital circuits for the purpose of investigating its size. Resultantly it is composed of a printed wiring board with the size of 15cm×15cm, and 26 integrated circuits. Half the printed wiring board is occupied by the memory unit to retain the weights, because the memory unit is composed of digital memories, A/D and D/A converters. Thus Folthret will become more compact when an appropriate CCD with large capacity is developed. Secondly with the Folthret realized by electric circuits, a learning experiment of various kinds of two-class pattern classifications has been conducted by using 20 random patterns for the purpose of evaluating its learning capability. As a result, it was revealed that Folthret can learn every two-class pattern classification mentioned above. Also the number of learning cycles etc. have been confirmed to be fairly in accordance with the case of the simulation using of a learning threshold-element model.