Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2510.05782 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915562719805440 |
|---|---|
| author | De la Jara, I. M. Rodriguez-Opazo, C. Teney, D. Ranasinghe, D. Abbasnejad, E. |
| author_facet | De la Jara, I. M. Rodriguez-Opazo, C. Teney, D. Ranasinghe, D. Abbasnejad, E. |
| contents | Out-of-distribution (OOD) detection is essential for reliably deploying machine learning models in the wild. Yet, most methods treat large pre-trained models as monolithic encoders and rely solely on their final-layer representations for detection. We challenge this wisdom. We reveal the \textit{intermediate layers} of pre-trained models, shaped by residual connections that subtly transform input projections, \textit{can} encode \textit{surprisingly rich and diverse signals} for detecting distributional shifts. Importantly, to exploit latent representation diversity across layers, we introduce an entropy-based criterion to \textit{automatically} identify layers offering the most complementary information in a training-free setting -- \textit{without access to OOD data}. We show that selectively incorporating these intermediate representations can increase the accuracy of OOD detection by up to \textbf{$10\%$} in far-OOD and over \textbf{$7\%$} in near-OOD benchmarks compared to state-of-the-art training-free methods across various model architectures and training objectives. Our findings reveal a new avenue for OOD detection research and uncover the impact of various training objectives and model architectures on confidence-based OOD detection methods. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_05782 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Mysteries of the Deep: Role of Intermediate Representations in Out of Distribution Detection De la Jara, I. M. Rodriguez-Opazo, C. Teney, D. Ranasinghe, D. Abbasnejad, E. Computer Vision and Pattern Recognition Out-of-distribution (OOD) detection is essential for reliably deploying machine learning models in the wild. Yet, most methods treat large pre-trained models as monolithic encoders and rely solely on their final-layer representations for detection. We challenge this wisdom. We reveal the \textit{intermediate layers} of pre-trained models, shaped by residual connections that subtly transform input projections, \textit{can} encode \textit{surprisingly rich and diverse signals} for detecting distributional shifts. Importantly, to exploit latent representation diversity across layers, we introduce an entropy-based criterion to \textit{automatically} identify layers offering the most complementary information in a training-free setting -- \textit{without access to OOD data}. We show that selectively incorporating these intermediate representations can increase the accuracy of OOD detection by up to \textbf{$10\%$} in far-OOD and over \textbf{$7\%$} in near-OOD benchmarks compared to state-of-the-art training-free methods across various model architectures and training objectives. Our findings reveal a new avenue for OOD detection research and uncover the impact of various training objectives and model architectures on confidence-based OOD detection methods. |
| title | Mysteries of the Deep: Role of Intermediate Representations in Out of Distribution Detection |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2510.05782 |