Abstract
Nowadays, the usage of electronic devices running artificial neural networks (ANNs)-based applications is spreading in our everyday life. Due to their outstanding computational capabilities, ANNs have become appealing solutions for safety-critical systems as well. Frequently, they are considered intrinsically robust and fault tolerant for being brain-inspired and redundant computing models. However, when ANNs are deployed on resource-constrained hardware devices, single physical faults may compromise the activity of multiple neurons. Therefore, it is crucial to assess the reliability of the entire neural computing system, including both the software and the hardware components. This article systematically addresses reliability concerns for ANNs running on multiprocessor system-on-a-chips (MPSoCs). It presents a methodology to assign resilience scores to individual neurons and, based on that, schedule the workload of an ANN on the target MPSoC so that critical neurons are neatly distributed among the available processing elements. This reliability-oriented methodology exploits an integer linear programming solver to find the optimal solution. Experimental results are given for three different convolutional neural networks trained on MNIST, SVHN, and CIFAR-10. We carried out a comprehensive assessment on an open-source artificial intelligence-based RISC-V MPSoC. The results show the reliability improvements of the proposed methodology against the traditional scheduling.
Highlights
In recent years, to face the growing complexity of emerging computing systems and algorithms, artificial intelligence (AI)-based solutions and, brain-inspired computing models have gained large interest in both industry and academia
As for the Random scenario, since we relied on a random choice of neurons to kill, the experiments were repeated 1000 times; we report in Figures 4–6 the average percentage obtained through the experiments
This paper provides a methodology to improve the reliability of a neural computing system running in a multi-core device
Summary
To face the growing complexity of emerging computing systems and algorithms, artificial intelligence (AI)-based solutions and, brain-inspired computing models have gained large interest in both industry and academia. Researchers have developed artificial models named artificial neural networks (ANNs) by imitating biological neurons and their functioning in the human brain. Since their origin [1], a huge number of studies have made progress in improving the theory behind brain-inspired computations to build highly complex artificial models, such as deep neural networks (DNNs). The human brain is a complex and fascinating system able to bear synapse or neuron faults and still keep working properly, thanks to its plastic ability to remodel, repair, and reorganize its neural functions [2]. It can be considered a unique block comprising two non-independent levels: Behavioral level : It includes the technology independent artificial neural network software model
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have