Aktive Sehsysteme mit biologienahen Neuronalen Netzen Abschlussbericht

Compared to other well known kinds of artificial neural nets, pulse-coded neural networks are much closer to the biological system and show some advantages for the use in active vision systems. Therefore, the goal of this project was to develop a special hardware for efficient simulation of large pulse-coded neural nets. Based on standard components like FPGAs and CPLDs we build a modular system called the 'FPGA-Accelerator' which takes advantage of the special characteristics of pulse-coded nets and allows the simulation of 131072 neurons with 16777216 connections. A network of several accelerator-systems allows simulation of even larger nets. The IO-ports of the accelerator are realized with standard transputer-links. This makes it easy to connect the system to various computer-systems. (orig.)SIGLEAvailable from TIB Hannover: F97B1389+a / FIZ - Fachinformationszzentrum Karlsruhe / TIB - Technische InformationsbibliothekBundesministerium fuer Bildung, Wissenschaft, Forschung und Technologie, Bonn (Germany)DEGerman