Fault Resilience

Fault resilience in artificial neural networks (ANNs) focuses on designing and analyzing systems that maintain functionality despite hardware failures or data corruption. Current research emphasizes developing methods to improve the robustness of various ANN architectures, including spiking neural networks (SNNs) and quantized neural networks (QNNs), using techniques like neuron splitting, data remapping, and graph neural networks (GNNs) for fault detection and diagnosis. This work is crucial for deploying ANNs in safety-critical applications like autonomous vehicles and smart grids, where reliability is paramount, and for advancing the trustworthiness of AI systems more broadly. The development of efficient fault-tolerance mechanisms is a key challenge driving ongoing research.