Title: Error Resilient and Adaptive Deep Learning Systems
Committee:
Dr. Chatterjee, Advisor
Dr. Aghazadeh, Chair
Dr. Hao
Abstract: The objective of proposed research is to develop error resilience and adaptive deep learning systems on various Artificial Neural Networks (ANNs) architecture and applications. This objective is motivated by accompanying errors and hardware failures of such deep learning systems due to its rapid growth of implementation on real world applications. These errors and hardware failures can be induced by permanent faults, transient faults, and parametric variations in both digital system-level and analog circuit-level. Key focuses are dedicated error detection and mitigation methodologies for faults and parametric variations considering architecture of network and characteristics of implemented hardware.