Fault-Resistant Algorithms for Single Layer Neural Networks

Abstract

Incremental extreme learning machine (IELM), convex incremental extreme learning machine (C-IELM) and other variants of extreme learning machine (ELM) algorithms provide low computational complexity techniques for training single layer feed-forward networks (SLFNs). However, the original IELM and C-IELM consider faultless network situations only. This paper investigates the performance of IELM and C-IELM under the multiplicative weight noise situation, where the input weights and the output weights are contaminated by noise. In addition, we propose two incremental fault tolerant algorithms, namely weight deviation tolerant-IELM (WDT-IELM) and weight deviation tolerant convex-IELM (WDTC-IELM). The performance of the two proposed algorithms is better than that of the two original ELM algorithms. Moreover, the convergence properties of the proposed algorithms are presented.