Adaptive Soft Error Protection for Neural Network Processing

2024-07-29

AI Summary

This work addresses the high computational overhead and inflexibility of static soft-error protection mechanisms in neural networks, which fail to adapt to input-dependent vulnerability variations. We first reveal and formally model the strong input dependence of soft-error vulnerability. To this end, we propose an input-aware dynamic vulnerability modeling and adaptive protection framework: a lightweight graph neural network (GNN) is designed for fine-grained, real-time vulnerability prediction; based on these predictions, selective redundancy and error-correcting codes are dynamically activated. Experiments across multiple datasets and network architectures demonstrate that our approach reduces average computational overhead by 42.12% while strictly preserving original reliability. Our core contribution is the establishment of the first input-driven paradigm for real-time soft-error vulnerability prediction, enabling joint optimization of accuracy, computational cost, and reliability.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Hardware-aware ML
• Natural Language Processing: Safety and Robustness

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Security and Privacy: Large-scale security measurements
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Mitigating soft errors in neural networks (NNs) often incurs significant computational overhead. Traditional methods mainly explored static vulnerability variations across NN components, employing selective protection to minimize costs. In contrast, this work reveals that NN vulnerability is also input-dependent, exhibiting dynamic variations at runtime. To this end, we propose a lightweight graph neural network (GNN) model capable of capturing input- and component-specific vulnerability to soft errors. This model facilitates runtime vulnerability prediction, enabling an adaptive protection strategy that dynamically adjusts to varying vulnerabilities. The approach complements classical fault-tolerant techniques by tailoring protection efforts based on real-time vulnerability assessments. Experimental results across diverse datasets and NNs demonstrate that our adaptive protection method achieves a 42.12% average reduction in computational overhead compared to prior static vulnerability-based approaches, without compromising reliability.

Problem

• Mitigate soft errors in neural networks.
• Reduce computational overhead in error protection.
• Adapt protection based on runtime vulnerability.

Innovation

• Adaptive GNN model
• Runtime vulnerability prediction
• Dynamic protection strategy