A generalized ABFT technique using a fault tolerant neural network

• Main Authors: A. Moosavienia, K. Mohammadi
• Format: Article
• Language: English
• Published: Iran University of Science and Technology 2005-01-01
• Series: Iranian Journal of Electrical and Electronic Engineering
• Subjects: fault tolerance; back propagation; MLP network; function approximation; ABFT; convolutional codes; majority logic decoding.
• Online Access: http://ijeee.iust.ac.ir/browse.php?a_code=A-10-3-35&slc_lang=en&sid=1
• ISSN: 1735-2827; 2383-3890

In this paper we first show that standard BP algorithm cannot yeild to a uniform information distribution over the neural network architecture. A measure of sensitivity is defined to evaluate fault tolerance of neural network and then we show that the sensitivity of a link is closely related to the amount of information passes through it. Based on this assumption, we prove that the distribution of output error caused by s-a-0 (stuck at 0) faults in a MLP network has a Gaussian distribution function. UDBP (Uniformly Distributed Back Propagation) algorithm is then introduced to minimize mean and variance of the output error. Simulation results show that UDBP has the least sensitivity and the highest fault tolerance among other algorithms such as WRTA, N-FTBP and ADP. Then a MLP neural network trained with UDBP, contributes in an Algorithm Based Fault Tolerant (ABFT) scheme to protect a nonlinear data process block. The neural network is trained to produce an all zero syndrome sequence in the absence of any faults. A systematic real convolution code guarantees that faults representing errors in the processed data will result in notable nonzero values in syndrome sequence. A majority logic decoder can easily detect and correct single faults by observing the syndrome sequence. Simulation results demonstrating the error detection and correction behavior against random s-a-0 faults are presented too.