Deep Stochastic Configuration Networks

In contrast to known randomized learning algorithms for single layer feed-forward neural networks (e.g., random vector functional-link networks), Stochastic Configuration Networks (SCNs) randomly assign the input weights and biases of the hidden nodes in the light of a supervisory mechanism, while the output weights are analytically evaluated in a constructive or selective manner.

Current experimental results indicate that SCNs outperform other randomized neural networks in terms of required human intervention, selection of the scope of random parameters, and fast learning and generalization. Deep sctochastic configuration networks (DeepSCNs) have been mathematically proved as universal approximators for continous nonlinear functions defined over compact sets. They can be constructed efficiently (much faster than other deep neural networks) and share many great features, such as learning representation and consistency property between learning and generalization.

This website collect some introductory material on DeepSCNs, most notably a brief selection of publications and some software to get started.

News

Deep stochastic configuration networks homepage is released on June 1, 2017.

The first release of the DeepSCNs website is available with initial code and publications.