Divide and Conquer: Object Co-occurrence Helps Mitigate Simplicity Bias in OOD Detection

Researchers propose OCO (Object Co-occurrence), a new out-of-distribution detection framework that leverages object co-occurrence patterns within images to improve the reliability of deep learning models. The method addresses simplicity bias by learning disentangled representations and using divide-and-conquer logic to distinguish near-OOD samples, achieving competitive results across multiple OOD detection benchmarks.

This research tackles a fundamental problem in machine learning reliability: detecting when input data falls outside a model's training distribution. Traditional OOD detection methods rely on entangled representations that often suffer from simplicity bias, where models learn superficial features rather than semantically meaningful patterns. The OCO framework introduces a paradigm shift by incorporating object co-occurrence patterns—how objects naturally appear together in real-world scenes—as a contextual signal for detection.

The work builds on established principles from cognitive science and human vision, where contextual understanding enhances scene recognition. By moving beyond single-object feature analysis, OCO creates a more robust detection mechanism that can identify near-OOD samples, which represent the most challenging detection scenario because they share visual similarities with in-distribution data. The divide-and-conquer approach adaptively segments patterns into three categories based on observed co-occurrence relationships in training data, enabling more nuanced discrimination.

For the AI development community, this research has practical implications for deploying neural networks in safety-critical applications where false negatives carry high costs. The method addresses both semantic shifts (different object types) and covariate shifts (different environmental contexts), improving model robustness across diverse scenarios. The public code release accelerates adoption and reproducibility.

Looking forward, this approach could influence how computer vision systems are evaluated and deployed, particularly in autonomous systems and medical imaging where OOD detection directly impacts reliability. Future work may extend co-occurrence modeling to other domains beyond vision or integrate it with existing uncertainty quantification methods.

• →OCO framework leverages object co-occurrence patterns to improve out-of-distribution detection accuracy, particularly for near-OOD samples.
• →The method addresses simplicity bias by learning disentangled representations guided by contextual relationships observed in training data.
• →Divide-and-conquer approach adaptively categorizes patterns into three scenarios based on object co-occurrence, enabling nuanced OOD discrimination.
• →Research demonstrates competitive performance across challenging OOD settings with capability to handle both semantic and covariate shifts.
• →Open-source code release facilitates adoption and enables integration into real-world deployment pipelines for safety-critical applications.