Exploiting the Tradeoff between Program Accuracy and Soft-error Resiliency Overhead for Machine Learning Workloads (1707.02589v1)

Abstract: To protect multicores from soft-error perturbations, resiliency schemes have been developed with high coverage but high power and performance overheads. Emerging safety-critical machine learning applications are increasingly being deployed on these platforms. Moreover, these systems are exposed to harsh environments, such as unmanned aerial vehicles (UAVs) and self-driving cars. Due to the unique structure and computational behavior of such applications, research has been done on relaxing their accuracy for performance benefits. We observe that not all transient errors affect program correctness, some errors only affect program accuracy, i.e., the program completes with certain acceptable deviations from error free outcome. This paper illustrates the idea of cross-layer soft-error resilience using machine learning workloads, where program accuracy is introduced as a tradeoff to deliver resilient yet efficient execution on futuristic large-scale multicores.