INTRODUCTION
NEURAL networks are computational models of the brain. These networks are good at solving problems for which a solutions seems easy to obtain for the brain, but requires a lot of efforts using standard algorithmic techniques. Examples of such problems are pattern recognition, perception, generalization and non-linear control. In the brain, all communication between neurons occurs using action potentials or spikes. In classical neural models these individual spikes are averaged out in time and all interaction is identified by the mean firing rate of the neurons.
Recently there has been an increasing interest in more complex models, which take the individual spikes into account. This sudden interest is catalyzed by the fact that these more realistic models are very well suited for hardware implementations. In addition they are computationally stronger than classic neural networks.
Spiking neural networks are better suited for hardware implementations due to two facts: inter-neuron communication consists of single bits and the neurons themselves are actually only weighed leaky integrators. Because only single bits of information need to be transmitted, a single wire is sufficient for connection between two neurons. Thereby the routing of neural interconnection on a 2D chip is implied. The neural model that is used is the ‘integrate-and-fire’ model. Neurons here are…………….So on ..........(download any of the following links to get complete paper presentation in word document)
No comments:
Post a Comment