Breaking Semantic Hegemony: Decoupling Principal and Residual Subspaces for Generalized OOD Detection
This addresses a critical limitation in OOD detection for AI safety by improving robustness against structurally distinct samples and sensor noise.
The paper tackles the Simplicity Paradox in OOD detection where models are sensitive to semantically subtle samples but blind to structurally distinct ones, attributing this to Semantic Hegemony in feature spaces. The proposed D-KNN method decouples semantic and structural components, achieving new SOTA results including reducing FPR95 from 31.3% to 2.3% and boosting AUROC from 79.7% to 94.9%.
While feature-based post-hoc methods have made significant strides in Out-of-Distribution (OOD) detection, we uncover a counter-intuitive Simplicity Paradox in existing state-of-the-art (SOTA) models: these models exhibit keen sensitivity in distinguishing semantically subtle OOD samples but suffer from severe Geometric Blindness when confronting structurally distinct yet semantically simple samples or high-frequency sensor noise. We attribute this phenomenon to Semantic Hegemony within the deep feature space and reveal its mathematical essence through the lens of Neural Collapse. Theoretical analysis demonstrates that the spectral concentration bias, induced by the high variance of the principal subspace, numerically masks the structural distribution shift signals that should be significant in the residual subspace. To address this issue, we propose D-KNN, a training-free, plug-and-play geometric decoupling framework. This method utilizes orthogonal decomposition to explicitly separate semantic components from structural residuals and introduces a dual-space calibration mechanism to reactivate the model's sensitivity to weak residual signals. Extensive experiments demonstrate that D-KNN effectively breaks Semantic Hegemony, establishing new SOTA performance on both CIFAR and ImageNet benchmarks. Notably, in resolving the Simplicity Paradox, it reduces the FPR95 from 31.3% to 2.3%; when addressing sensor failures such as Gaussian noise, it boosts the detection performance (AUROC) from a baseline of 79.7% to 94.9%.