A Simple Fix to Mahalanobis Distance for Improving Near-OOD Detection

2021/06/16 by Jie Ren, Ren, Jie, Stanislav Fort +9 · 24 citations
Computer Science · #Adversarial Robustness in Machine Learning #Domain Adaptation and Few-Shot Learning #Advanced Neural Network Applications

paper · pdf · doi:10.48550/arxiv.2106.09022

Abstract

Mahalanobis distance (MD) is a simple and popular post-processing method for detecting out-of-distribution (OOD) inputs in neural networks. We analyze its failure modes for near-OOD detection and propose a simple fix called relative Mahalanobis distance (RMD) which improves performance and is more robust to hyperparameter choice. On a wide selection of challenging vision, language, and biology OOD benchmarks (CIFAR-100 vs CIFAR-10, CLINC OOD intent detection, Genomics OOD), we show that RMD meaningfully improves upon MD performance (by up to 15% AUROC on genomics OOD).

Citations

• A Simple Unified Framework for Detecting Out-of-Distribution Samples and Adversarial Attacks
• Learning Transferable Visual Models From Natural Language Supervision

Cited by

• RankOOD -- Class Ranking-based Out-of-Distribution Detection
• ClusterMine: Robust Label-Free Visual Out-Of-Distribution Detection via Concept Mining from Text Corpora
• Generalized Out-of-Distribution Detection: A Survey
• Beyond Binary Out-of-Distribution Detection: Characterizing Distributional Shifts with Multi-Statistic Diffusion Trajectories
• Learning After Model Deployment
• GOOD: Training-Free Guided Diffusion Sampling for Out-of-Distribution Detection
• Benchmarking Out-of-Distribution Detection for Plankton Recognition: A Systematic Evaluation of Advanced Methods in Marine Ecological Monitoring
• Dissecting Mahalanobis: How Feature Geometry and Normalization Shape OOD Detection
• Towards Robust Artificial Intelligence: Self-Supervised Learning Approach for Out-of-Distribution Detection
• Medix: Out-of-Distribution Detection from Unlabeled Wild Data via Robust Gradient Statistics
• ICONIC-444: A 3.1-Million-Image Dataset for OOD Detection Research
• Integrating Contextual Embeddings into Evaluation of Expressive MIDI Piano Performances
• PGTuner: An Efficient Framework for Automatic and Transferable Configuration Tuning of Proximity Graphs
• An Empirical Analysis of VLM-based OOD Detection: Mechanisms, Advantages, and Sensitivity
• DCV-ROOD Evaluation Framework: Dual Cross-Validation for Robust Out-of-Distribution Detection
• Activation Subspaces for Out-of-Distribution Detection
• Probability Density from Latent Diffusion Models for Out-of-Distribution Detection
• Rapidly Adapting to New Voice Spoofing: Few-Shot Detection of Synthesized Speech Under Distribution Shifts
• Pseudo-label Induced Subspace Representation Learning for Robust Out-of-Distribution Detection
• BOOD: Boundary-based Out-Of-Distribution Data Generation
• Zero-Shot Image Anomaly Detection Using Generative Foundation Models
• COOkeD: Ensemble-based OOD detection in the era of zero-shot CLIP
• Knowledge Regularized Negative Feature Tuning of Vision-Language Models for Out-of-Distribution Detection
• Synthesizing Near-Boundary OOD Samples for Out-of-Distribution Detection

Related

• Why is the Mahalanobis Distance Effective for Anomaly Detection?
• On the use of Mahalanobis distance for out-of-distribution detection with neural networks for medical imaging
• Mahalanobis++: Improving OOD Detection via Feature Normalization
• Adversarial vulnerability of powerful near out-of-distribution detection
• Dissecting Mahalanobis: How Feature Geometry and Normalization Shape OOD Detection