Divisive Decisions: Improving Salience-Based Training for Generalization in Binary Classification Tasks

2025-07-22

AI Summary

Existing saliency-guided training solely aligns ground-truth class CAMs with human-annotated saliency maps, neglecting discriminative information from erroneous-class CAMs. To address insufficient generalization in binary classification tasks, this work introduces erroneous-class CAMs for the first time and proposes a “True-vs.-False Class Activation Difference” mechanism. By modeling the contrastive responses of true and false class CAMs within human-identified critical feature regions, we devise three novel fused training strategies and a post-hoc explainability analysis tool. Our approach integrates CAMs, saliency-aware loss, and contrastive learning, supervised by human-annotated saliency maps. Extensive experiments on face forgery detection, biometric attack identification, and chest X-ray abnormality classification demonstrate that our method significantly outperforms state-of-the-art saliency-guided approaches, achieving superior generalization and enhanced focus on task-critical features.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Calibration & Uncertainty Quantification
• Humans and AI: Explainable AI (XAI) for Human Understanding

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Existing saliency-guided training approaches improve model generalization by incorporating a loss term that compares the model's class activation map (CAM) for a sample's true-class ({it i.e.}, correct-label class) against a human reference saliency map. However, prior work has ignored the false-class CAM(s), that is the model's saliency obtained for incorrect-label class. We hypothesize that in binary tasks the true and false CAMs should diverge on the important classification features identified by humans (and reflected in human saliency maps). We use this hypothesis to motivate three new saliency-guided training methods incorporating both true- and false-class model's CAM into the training strategy and a novel post-hoc tool for identifying important features. We evaluate all introduced methods on several diverse binary close-set and open-set classification tasks, including synthetic face detection, biometric presentation attack detection, and classification of anomalies in chest X-ray scans, and find that the proposed methods improve generalization capabilities of deep learning models over traditional (true-class CAM only) saliency-guided training approaches. We offer source codes and model weightsfootnote{GitHub repository link removed to preserve anonymity} to support reproducible research.

Problem

• Improving model generalization using true and false class activation maps
• Enhancing binary classification with saliency-guided training methods
• Evaluating methods on diverse tasks like face detection and X-ray analysis

Innovation

• Incorporates true and false class CAMs
• Introduces three new saliency-guided methods
• Improves model generalization in binary tasks