Phenomenon of tolerance to damage in artificial neural networks
Ryszard Tadeusiewicz, Izabela Figura · Computer Methods in Materials Science. · 2011
Neural networks are computer tools with many advantages. They are powerful because of possibility of complex nonlinear systems modeling and they are user-friendly because of learning abilities. But neural networks have additional advantages. One of them is increased tolerance to damages. Technological systems are mainly not resistive for damages. Microprocessor if is even a little damaged – just doesn’t work. In contrary to this biological systems, especially brain are almost insensitive for damages – many cell can be dead or can have many malfunctions – but the brain as a whole can functioning properly. The research presented in the paper is dedicated for discovery, if neural networks, as a models of natural neural system elements, can show tolerance to damages. This assumption was experimentally proven and results of simulations showing such resistance, are presented in the paper.