Abstract:

When using sparse neural networks in practice, it is hard to choose a proper connection rate. In this work, based on new discoveries in the brain science, two new learning algorithms are developed which change the network’s connection structure at the time of learning, thus an accurate connection rate is not needed. Sparse neural networks reduce coupling among inputs so that fewer connections are needed to meet the fitting requirement. Simulation results show that the new algorithms are effective.

Key words: sparse neural network, generalization, learning algorithm, connectivily