New HPM detector boosts OOD detection with simpler feature geometry

Researchers propose a new post-hoc detector called Hyperspherical Pooled Mahalanobis (HPM) for long-tailed out-of-distribution (LT-OOD) detection. This method normalizes features and uses a pooled, ridge-regularized metric instead of class-specific covariances. Experiments on CIFAR-LT and ImageNet-100-LT datasets show HPM significantly improves detection accuracy, achieving state-of-the-art Log Efficiency Score on CIFAR-100-LT while maintaining high AUROC with reduced training costs. AI

IMPACT Introduces a more efficient method for detecting out-of-distribution data, potentially improving model robustness in real-world applications.

RANK_REASON The cluster contains an academic paper detailing a new method for a specific machine learning task. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• CIFAR-LT
• Hyperspherical Pooled Mahalanobis (HPM)
• ImageNet-100-LT
• Prior-Calibrated ERM (PC-ERM)

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →