Presentation
Late Breaking Results: RQ-DNN: Reliable Quantization for Fault-tolerant Deep Neural Network
DescriptionDeep Neural Networks (DNNs) are deployed in many safety-critical applications such as autonomous vehicles. However, recent research has shown that faults in memory can cause a significant accuracy drop of DNN. This paper proposes a novel method which performs a model quantization with remarkably improving the fault-tolerance of the model and can be incorporated with other hardware approaches such as Error Correcting Code. The proposed method improves fault-tolerance by modifying weight distributions and by applying a novel clipping function. The proposed method enhances the fault-tolerance of DNN which can tolerate 1803x higher bit error rates than the conventional method.
Event Type
Late Breaking Results Poster
TimeWednesday, July 12th6:00pm - 7:00pm PDT
LocationLevel 2 Lobby
AI
Autonomous Systems
Cloud
Design
EDA
Embedded Systems
IP
Security