DAVE, MOHIT KUMAR

Abstract

The aim is to build a simple model of neural network that have biological features ie. the network is biologically plausible and that can search for correct output automatically. Here the purpose is not to build a complex model as of brain, but to study the principles of learning. Therefore a very simplified model is used. In this work a layered neural net with the following properties is studied: • It has adaptable weights. • The threshold potential is fixed. A feedback signal is used for this neural net, it is a binary signal, whose value depend on whether the output of the network for its input is wrong or right. According to some assumptions based on some biological theories, it comes out that two kind of learning are possible: • One is Hebbian • Other one is Anti–Hebbian What the Hebbian learning does is that it keep knowledge of input-output relations, so it occur when the output of the network is same as expected. While on the other hand Anti– Hebbian learning chan...