A Contrastive Teacher-Student Framework for Novelty Detection under Style Shifts

2025-01-28

AI Summary

Environmental style shifts degrade novelty detection performance, as models trained without out-of-distribution (OOD) samples conflate stylistic and semantic features. Method: We propose a style-aware auxiliary OOD data construction mechanism that synthesizes samples with style similarity but semantic dissimilarity; further, we design a core-feature-guided contrastive teacher-student distillation framework that explicitly disentangles semantic from style features via task-driven knowledge transfer. Contribution/Results: This is the first work to jointly integrate style-controllable augmentation, contrastive learning, and knowledge distillation for robust novelty detection. Evaluated on multiple synthetic and real-world benchmarks, our method consistently outperforms nine state-of-the-art approaches, achieving an average 12.6% AUROC improvement under style shift. It effectively mitigates reliance on style shortcuts and enhances cross-style generalization.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Adversarial Learning & Robustness
• Data Mining & Knowledge Management: Anomaly/Outlier Detection

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling

Abstract

There have been several efforts to improve Novelty Detection (ND) performance. However, ND methods often suffer significant performance drops under minor distribution shifts caused by changes in the environment, known as style shifts. This challenge arises from the ND setup, where the absence of out-of-distribution (OOD) samples during training causes the detector to be biased toward the dominant style features in the in-distribution (ID) data. As a result, the model mistakenly learns to correlate style with core features, using this shortcut for detection. Robust ND is crucial for real-world applications like autonomous driving and medical imaging, where test samples may have different styles than the training data. Motivated by this, we propose a robust ND method that crafts an auxiliary OOD set with style features similar to the ID set but with different core features. Then, a task-based knowledge distillation strategy is utilized to distinguish core features from style features and help our model rely on core features for discriminating crafted OOD and ID sets. We verified the effectiveness of our method through extensive experimental evaluations on several datasets, including synthetic and real-world benchmarks, against nine different ND methods.

Problem

• Novelty Detection
• Environmental Variability
• Style-dependent Bias

Innovation

• Teacher-Student Contrastive Framework
• Style-Invariant Novelty Detection
• Adaptive Teaching Strategy