Online Training for Open Faulty RBF Networks

• Published in: Neural processing letters Vol. 42; no. 2; pp. 397 - 416
• Main Authors: Xiao, Yi, Feng, Ruibin, Leung, Chi Sing, Sum, Pui Fai
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.10.2015; Springer Nature B.V
• Subjects: Adaptive learning; Algorithms; Artificial Intelligence; Brain; Complex Systems; Computational Intelligence; Computer Science; Distance learning; Fault tolerance; Machine learning; Neural networks; Online instruction; Random variables; Regularization; Online Mode; Learning Cycle; Weight Decay; Convergent Condition; Radial Basis Function
• ISSN: 1370-4621; 1573-773X
• DOI: 10.1007/s11063-014-9363-8

Recently, a batch mode learning algorithm, namely optimal open weight fault regularization (OOWFR), was developed for handling the open fault situation. In terms of the Kullback–Leibler divergence, this batch mode learning algorithm is optimal. However, the main disadvantage of this batch mode learning algorithm is that it requires to store the entire input–output history. Therefore, the memory consumption is a problem when the number of training samples is large. In this paper, we present an online version for the OOWFR algorithm. We consider two learning rate cases, fixed learning rate and adaptive learning rate. We present the convergent conditions for these two cases. Simulation results show that the performance of the proposed online mode learning algorithm is better than that of other online mode learning algorithms. Also, the performance of the proposed algorithm is close to that of the batch mode OOWFR algorithm.