Neural networks and expectation of VLSI implementation

2.1 Transformation Neural Network Computational neuroscience and its application to The most fundamental motivation for exploring neuengineering “neurocomputing” have been a subject ral networks as a guide to future information processof great interest for the past several years. Research ing machines comes from the fact that we do have of neurocomputing is expected to lead to developits most fascinating realization as the human brain. ment of massively parallel network systems based Neural networks have at least two remarkable charon neural network models: neurocompufer. There acteristics in contrast with the present von Neumann seems to be several reasons for the recent resurgence type computer. First it h a s the capability of learning of interest in neural network as an information prowith use of plastic changes of synaptic connections cessing machine. (i) Improvement of computer fabetween its computing elements neurons. Second it cility as a tool for simulating neural network modsolves computational problems by cooperative operels. (ii) Increasing feasibility of hardware implemenation of great number of neurons (10” in-the brain). tation of neural network models by analogue VLSI [1,2,3], optoelectronic devices [4] etc. (iii) Steady deAccording to the above two features, many Of the velopment of experimental neuroscience. (iV) connetwork models, which were proposed t o account for sidetable amount of fundamental principles and neubrain functions such pattern recognition or memral network models accumulated during past 25 years ory in a somewhat abstract can be classiresearch of biological information processing. (v) AI fied into the two The research, which aims at the similar intelligence as huis called as “transformation” neural network model