Cluster Random Weight Change for Neural Network Learning

In this work a new stochastic learning algorithm, called Cluster Random Weight Change (CRWC) is proposed. The network is subdivided in a fixed number of neurons to form some clusters of neu- rons and learning is applied to one cluster at a time. The experimental results show an improvement in learning accuracy and a less critical dependence on noise amplitude and annealing parameters. A great ad- vantage of this algorithm is the reduced chip area occupancy required, as it has been shown in our hardware implementation.