Causality-based Neural Network Repair

TL;DR

This paper introduces CARE, a causality-based method for repairing neural networks to improve fairness, safety, and security by minimally adjusting parameters based on fault localization, demonstrating high effectiveness across various tasks.

Contribution

The paper presents a novel causality-based neural network repair technique that localizes faults and repairs networks while preserving accuracy, addressing safety, fairness, and security concerns.

Findings

CARE effectively repairs neural networks for fairness, safety, and backdoor removal.

CARE reduces attack success rate from over 98% to below 1%.

Fairness improves by 61.91% on average.

Abstract

Neural networks have had discernible achievements in a wide range of applications. The wide-spread adoption also raises the concern of their dependability and reliability. Similar to traditional decision-making programs, neural networks can have defects that need to be repaired. The defects may cause unsafe behaviors, raise security concerns or unjust societal impacts. In this work, we address the problem of repairing a neural network for desirable properties such as fairness and the absence of backdoor. The goal is to construct a neural network that satisfies the property by (minimally) adjusting the given neural network's parameters (i.e., weights). Specifically, we propose CARE (\textbf{CA}usality-based \textbf{RE}pair), a causality-based neural network repair technique that 1) performs causality-based fault localization to identify the `guilty' neurons and 2) optimizes the parameters of the identified neurons to reduce the misbehavior. We have empirically evaluated CARE on various tasks such as backdoor removal, neural network repair for fairness and safety properties. Our experiment results show that CARE is able to repair all neural networks efficiently and effectively. For fairness repair tasks, CARE successfully improves fairness by $61.91\%$ on average. For backdoor removal tasks, CARE reduces the attack success rate from over $98\%$ to less than $1\%$. For safety property repair tasks, CARE reduces the property violation rate to less than $1\%$. Results also show that thanks to the causality-based fault localization, CARE's repair focuses on the misbehavior and preserves the accuracy of the neural networks.