Hardware Implementation of Neural Networks

In this presentation I will talk about possible hardware
implementations of Neural Networks. Neural Networks are simple
processing elements that are able to do complex computations as
pattern recognition, system identification and control, data mining,
decision making or time series prediction. Neural Networks are
traditionally implemented using software tools while some specific
programs such as MATLAB provide special toolboxes for Neural Network
training and execution. Nevertheless hardware implementations take the
advantage of the inherent parallelism of NNs and also are much faster
and reliable if compared to software solutions. In this talk I will
show some hardware alternatives to implement Neural Networks. I will
also introduce the concept of stochastic computing that is an
essential tool to implement compact networks. Special attention
would be given to Spiking Neural Networks that mimic the behavior or
real biological neurons. A simple implementation of Spiking Neural
Networks will be presented along with an efficient learning
methodology based on the use of Genetic Algorithms.