Class based Influence Functions for Error Detection

2023/05/02 by Thang Nguyen-Duc, Nguyen-Duc, Thang, Hoang Thanh-Tung +11 · 2 citations
Computer Science · #Advanced Malware Detection Techniques #Anomaly Detection Techniques and Applications #Computation and Language (cs.CL) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Network Security and Intrusion Detection

paper · pdf · doi:10.48550/arxiv.2305.01384

openalex publication_date 2023/05/02 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28

Abstract

Influence functions (IFs) are a powerful tool for detecting anomalous examples in large scale datasets. However, they are unstable when applied to deep networks. In this paper, we provide an explanation for the instability of IFs and develop a solution to this problem. We show that IFs are unreliable when the two data points belong to two different classes. Our solution leverages class information to improve the stability of IFs. Extensive experiments show that our modification significantly improves the performance and stability of IFs while incurring no additional computational cost.

Cited by

• Detecting and Rectifying Noisy Labels: A Similarity-based Approach
• A Comparative Analysis of Influence Signals for Data Debugging

Related

• Influence Functions in Deep Learning Are Fragile
• Cross-Loss Influence Functions to Explain Deep Network Representations
• Learning Influence Functions from Incomplete Observations
• Theoretical and Practical Perspectives on what Influence Functions Do
• Towards Robust Influence Functions with Flat Validation Minima