Cost-Effective Fault Tolerance for CNNs Using Parameter Vulnerability Based Hardening and Pruning

TL;DR

This paper presents a hardware-agnostic, model-level fault tolerance method for CNNs that combines parameter vulnerability analysis with pruning to achieve high resilience and efficiency with minimal accuracy loss.

Contribution

It introduces a novel parameter vulnerability-based hardening and pruning approach that enhances fault tolerance in CNNs without hardware modifications.

Findings

Fault resilience comparable to TMR with lower overhead

Pruned CNNs are up to 24% faster after hardening

Proposed method outperforms conventional pruning techniques

Abstract

Convolutional Neural Networks (CNNs) have become integral in safety-critical applications, thus raising concerns about their fault tolerance. Conventional hardware-dependent fault tolerance methods, such as Triple Modular Redundancy (TMR), are computationally expensive, imposing a remarkable overhead on CNNs. Whereas fault tolerance techniques can be applied either at the hardware level or at the model levels, the latter provides more flexibility without sacrificing generality. This paper introduces a model-level hardening approach for CNNs by integrating error correction directly into the neural networks. The approach is hardware-agnostic and does not require any changes to the underlying accelerator device. Analyzing the vulnerability of parameters enables the duplication of selective filters/neurons so that their output channels are effectively corrected with an efficient and robust correction layer. The proposed method demonstrates fault resilience nearly equivalent to TMR-based correction but with significantly reduced overhead. Nevertheless, there exists an inherent overhead to the baseline CNNs. To tackle this issue, a cost-effective parameter vulnerability based pruning technique is proposed that outperforms the conventional pruning method, yielding smaller networks with a negligible accuracy loss. Remarkably, the hardened pruned CNNs perform up to 24\% faster than the hardened un-pruned ones.