Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Transient Fault Tolerant Semantic Segmentation for Autonomous Driving

Leonardo Iurada, Niccolò Cavagnero, Fernando Fernandes Dos Santos, Giuseppe Averta, Paolo Rech, Tatiana Tommasi

TL;DR

This work evaluates existing hardware fault models both in terms of accuracy and uncertainty and introduces ReLUMax, a novel simple activation function designed to enhance resilience against transient faults and integrates seamlessly into existing architectures without time overhead.

Abstract

Deep learning models are crucial for autonomous vehicle perception, but their reliability is challenged by algorithmic limitations and hardware faults. We address the latter by examining fault-tolerance in semantic segmentation models. Using established hardware fault models, we evaluate existing hardening techniques both in terms of accuracy and uncertainty and introduce ReLUMax, a novel simple activation function designed to enhance resilience against transient faults. ReLUMax integrates seamlessly into existing architectures without time overhead. Our experiments demonstrate that ReLUMax effectively improves robustness, preserving performance and boosting prediction confidence, thus contributing to the development of reliable autonomous driving systems.

Transient Fault Tolerant Semantic Segmentation for Autonomous Driving

TL;DR

This work evaluates existing hardware fault models both in terms of accuracy and uncertainty and introduces ReLUMax, a novel simple activation function designed to enhance resilience against transient faults and integrates seamlessly into existing architectures without time overhead.

Abstract

Deep learning models are crucial for autonomous vehicle perception, but their reliability is challenged by algorithmic limitations and hardware faults. We address the latter by examining fault-tolerance in semantic segmentation models. Using established hardware fault models, we evaluate existing hardening techniques both in terms of accuracy and uncertainty and introduce ReLUMax, a novel simple activation function designed to enhance resilience against transient faults. ReLUMax integrates seamlessly into existing architectures without time overhead. Our experiments demonstrate that ReLUMax effectively improves robustness, preserving performance and boosting prediction confidence, thus contributing to the development of reliable autonomous driving systems.
Paper Structure (9 sections, 3 equations, 2 figures, 2 tables)

This paper contains 9 sections, 3 equations, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Method
  4. Experiments
  5. Experimental Setting
  6. Semantic Segmentation Results
  7. Qualitative Effect of Fault Injections
  8. Uncertainty Analysis of Hardened Models
  9. Conclusions

Figures (2)

  • Figure 1: Semantic segmentation models may experience catastrophic output corruption under transient faults, rendering them unsafe for critical applications (No Hardening). To address this limitation, we propose a novel approach for automatically hardening activation functions at training time, without incurring in any additional cost (ReLUMax). Our method ensures robustness against transient faults, mitigating severe corruptions and significantly improving the system's trustworthiness.
  • Figure 2: Segmentation maps predicted by the methods under assessment, when simulating injections on the validation split of the Cityscapes dataset. Each color represents a pixel-level annotation of the corresponding semantic class. The choice of the example to visualize is based on the worst recorded mean Intersection over Union (mIoU) when no hardening is introduced. For fairness of comparison, all the methods are injected deterministically in the same way, according to the simulated fault model provided by cavagneroIOLTS2022.