Attribution-Guided Model Rectification of Unreliable Neural Network Behaviors

TL;DR

This paper introduces an attribution-guided model rectification framework using rank-one model editing to efficiently correct unreliable behaviors in neural networks, reducing data requirements and manual effort.

Contribution

It proposes a novel attribution-guided layer localization method and a formulation that corrects unreliabilities while preserving overall model performance.

Findings

Effective correction of neural Trojans, spurious correlations, and feature leakage.

Achieves model rectification with as few as one cleansed sample.

Reduces reliance on large-scale data cleaning and retraining.

Abstract

The performance of neural network models deteriorates due to their unreliable behavior on non-robust features of corrupted samples. Owing to their opaque nature, rectifying models to address this problem often necessitates arduous data cleaning and model retraining, resulting in huge computational and manual overhead. In this work, we leverage rank-one model editing to establish an attribution-guided model rectification framework that effectively locates and corrects model unreliable behaviors. We first distinguish our rectification setting from existing model editing, yielding a formulation that corrects unreliable behavior while preserving model performance and reducing reliance on large budgets of cleansed samples. We further reveal a bottleneck of model rectifying arising from heterogeneous editability across layers. To target the primary source of misbehavior, we introduce an attribution-guided layer localization method that quantifies layer-wise editability and identifies the layer most responsible for unreliabilities. Extensive experiments demonstrate the effectiveness of our method in correcting unreliabilities observed for neural Trojans, spurious correlations and feature leakage. Our method shows remarkable performance by achieving its editing objective with as few as a single cleansed sample, which makes it appealing for practice.