🤖 AI Summary
This work addresses the trade-off between scale invariance and hierarchical representational capacity in unsupervised out-of-distribution (OOD) detection by proposing a rigorous post-hoc framework that requires neither additional data nor model modifications. The method automatically selects discriminative intermediate layers exhibiting significant semantic compression via an entropy density descent criterion and integrates them through Ledoit-Wolf regularized covariance estimation, achieving scale-invariant and robust multi-layer feature fusion. A Top-K gating mechanism combined with Mahalanobis distance scoring is then employed within a unified feature space to produce stable OOD scores. Evaluated across diverse architectures, the approach demonstrates consistently strong performance on both near- and far-OOD samples, significantly outperforming existing unsupervised methods.
📝 Abstract
We introduce MM++ (Multilayer Mahalanobis++), a fully unsupervised, strictly post-hoc, and scale-invariant framework for Out-of-Distribution (OOD) detection. To address the trade-off between scale invariance and hierarchical expressivity, MM++ constructs a principled joint feature space. It first identifies discriminative intermediate layers by measuring entropy density drops, which mark the boundaries of sharp semantic compression. By fusing these selected layers with the terminal representation, the framework captures latent cross-layer correlations while mitigating early-layer noise. Crucially, a Ledoit-Wolf regularized tied covariance matrix stabilizes this unified space, enabling reliable distance estimation. Requiring no auxiliary OOD data, classifier fine-tuning, or architectural modifications, MM++ delivers robust performance across distinct architectures for both near- and far-OOD detection.