Fault tolerant broad learning system

Abstract

The broad learning system (BLS) approach provides low computational complexity solutions for training flat structure feedforward networks. However, many BLS algorithms deal with the faultless situation only. This paper addresses the fault tolerant ability of BLS networks. We call our approach fault tolerant BLS (FTBLS). First, we develop a fault tolerant objective function for BLS. Based on the developed objective function, we develop a training algorithm to construct a BLS network. The simulation results show that our proposed FTBLS is much better than the classical BLS.