Enhancing The Reliability of Out-of-distribution Image Detection in Neural Networks

Abstract

We consider the problem of detecting out-of-distribution images in neural networks. We propose ODIN, a simple and effective method that does not require any change to a pre-trained neural network. Our method is based on the observation that using temperature scaling and adding small perturbations to the input can separate the softmax score distributions between in- and out-of-distribution images, allowing for more effective detection. We show in a series of experiments that ODIN is compatible with diverse network architectures and datasets. It consistently outperforms the baseline approach by a large margin, establishing a new state-of-the-art performance on this task. For example, ODIN reduces the false positive rate from the baseline 34.7% to 4.3% on the DenseNet (applied to CIFAR-10) when the true positive rate is 95%.

ICLR 2018

References in corpus (5)

• Distilling the Knowledge in a Neural Network
• On Calibration of Modern Neural Networks
• Convolutional Neural Networks for Sentence Classification
• Regularizing Neural Networks by Penalizing Confident Output Distributions
• TurkerGaze: Crowdsourcing Saliency with Webcam based Eye Tracking

Cited by in corpus (120)

• FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence
• Aleatoric and Epistemic Uncertainty in Machine Learning: An Introduction to Concepts and Methods
• Deep Multi-modal Object Detection and Semantic Segmentation for Autonomous Driving: Datasets, Methods, and Challenges
• Adversarial Reciprocal Points Learning for Open Set Recognition
• Learning Open Set Network with Discriminative Reciprocal Points
• The Fishyscapes Benchmark: Measuring Blind Spots in Semantic Segmentation
• Zero-Shot Out-of-Distribution Detection Based on the Pre-trained Model CLIP
• Training Confidence-calibrated Classifiers for Detecting Out-of-Distribution Samples
• Exploring the Limits of Out-of-Distribution Detection
• Outlier Exposure with Confidence Control for Out-of-Distribution Detection
• The Conditional Entropy Bottleneck
• Deep Open Intent Classification with Adaptive Decision Boundary
• Testing for Outliers with Conformal p-values
• How to Certify Machine Learning Based Safety-critical Systems? A Systematic Literature Review
• The Pitfalls of Simplicity Bias in Neural Networks
• Your Classifier is Secretly an Energy Based Model and You Should Treat it Like One
• Adversarial Attack and Defense on Point Sets
• Contrastive Out-of-Distribution Detection for Pretrained Transformers
• Uncertainty in the Variational Information Bottleneck
• Embrace Limited and Imperfect Training Datasets: Opportunities and Challenges in Plant Disease Recognition Using Deep Learning
• Inspect, Understand, Overcome: A Survey of Practical Methods for AI Safety
• Gradient Starvation: A Learning Proclivity in Neural Networks
• A Post-processing Method for Detecting Unknown Intent of Dialogue System via Pre-trained Deep Neural Network Classifier
• Unified Probabilistic Deep Continual Learning through Generative Replay and Open Set Recognition
• Unsupervised Outlier Detection using Memory and Contrastive Learning
• The Functional Neural Process
• Entropic Out-of-Distribution Detection: Seamless Detection of Unknown Examples
• TEXTOIR: An Integrated and Visualized Platform for Text Open Intent Recognition
• Learning Discriminative Representations and Decision Boundaries for Open Intent Detection
• Outside the Box: Abstraction-Based Monitoring of Neural Networks
• AI Research Considerations for Human Existential Safety (ARCHES)
• Out-of-Distribution Detection using Multiple Semantic Label Representations
• OpenMatch: Open-set Consistency Regularization for Semi-supervised Learning with Outliers
• Unsupervised in-distribution anomaly detection of new physics through conditional density estimation
• Incorporating Experts' Judgment into Machine Learning Models
• Bayesian Variational Autoencoders for Unsupervised Out-of-Distribution Detection
• Attended Temperature Scaling: A Practical Approach for Calibrating Deep Neural Networks
• Unsupervised Out-of-Distribution Detection with Batch Normalization
• Understanding Open-Set Recognition by Jacobian Norm and Inter-Class Separation
• SPECTRE: Defending Against Backdoor Attacks Using Robust Statistics
• Diversity inducing Information Bottleneck in Model Ensembles
• Deep pNML: Predictive Normalized Maximum Likelihood for Deep Neural Networks
• SoK: Machine Learning Governance
• Unified Classification and Rejection: A One-versus-All Framework
• Are all outliers alike? On Understanding the Diversity of Outliers for Detecting OODs
• Sketching Curvature for Efficient Out-of-Distribution Detection for Deep Neural Networks
• Unsupervised Anomaly Detection with Adversarial Mirrored AutoEncoders
• Further Analysis of Outlier Detection with Deep Generative Models
• Open Set Medical Diagnosis
• A Unified Benchmark for the Unknown Detection Capability of Deep Neural Networks
• One Versus all for deep Neural Network Incertitude (OVNNI) quantification
• Accelerating Monte Carlo Bayesian Inference via Approximating Predictive Uncertainty over Simplex
• Detecting OODs as datapoints with High Uncertainty
• OneFlow: One-class flow for anomaly detection based on a minimal volume region
• Revisiting the Evaluation of Uncertainty Estimation and Its Application to Explore Model Complexity-Uncertainty Trade-Off
• Controlling Over-generalization and its Effect on Adversarial Examples Generation and Detection
• Extraction of Complex DNN Models: Real Threat or Boogeyman?
• On the Role of Dataset Quality and Heterogeneity in Model Confidence
• Open-World Lifelong Graph Learning
• Memory-Efficient Semi-Supervised Continual Learning: The World is its Own Replay Buffer
• What is Wrong with One-Class Anomaly Detection?
• Elsa: Energy-based learning for semi-supervised anomaly detection
• Efficacy of Pixel-Level OOD Detection for Semantic Segmentation
• Information-Theoretic Visual Explanation for Black-Box Classifiers
• Domain segmentation and adjustment for generalized zero-shot learning
• Uncertainty Quantification for Deep Context-Aware Mobile Activity Recognition and Unknown Context Discovery
• Pixel-wise Energy-biased Abstention Learning for Anomaly Segmentation on Complex Urban Driving Scenes
• Few-Shot Open-Set Recognition using Meta-Learning
• $k$Folden: $k$-Fold Ensemble for Out-Of-Distribution Detection
• Evaluation of Various Open-Set Medical Imaging Tasks with Deep Neural Networks
• Modeling Token-level Uncertainty to Learn Unknown Concepts in SLU via Calibrated Dirichlet Prior RNN
• A t-distribution based operator for enhancing out of distribution robustness of neural network classifiers
• Accelerating Domain-Aware Electron Microscopy Analysis Using Deep Learning Models with Synthetic Data and Image-Wide Confidence Scoring
• Exploring Covariate and Concept Shift for Detection and Calibration of Out-of-Distribution Data
• Unsupervised Temperature Scaling: An Unsupervised Post-Processing Calibration Method of Deep Networks
• Fine-grained Uncertainty Modeling in Neural Networks
• Synthetic Unknown Class Learning for Learning Unknowns
• On the Out-of-distribution Generalization of Probabilistic Image Modelling
• Adversarial TCAV -- Robust and Effective Interpretation of Intermediate Layers in Neural Networks
• Rainproof: An Umbrella To Shield Text Generators From Out-Of-Distribution Data
• UNO: Uncertainty-aware Noisy-Or Multimodal Fusion for Unanticipated Input Degradation
• An Alarm System For Segmentation Algorithm Based On Shape Model
• Analysis of skin lesion images with deep learning
• Semi-Supervised Class Discovery
• Anomaly Detection by Recombining Gated Unsupervised Experts
• Estimating Predictive Uncertainty Under Program Data Distribution Shift
• CVAD: A generic medical anomaly detector based on Cascade VAE
• Out-of-Distribution Detection for Adaptive Computer Vision
• DeepStreamCE: A Streaming Approach to Concept Evolution Detection in Deep Neural Networks
• Unsupervised Class-Incremental Learning Through Confusion
• Toward Scalable and Unified Example-based Explanation and Outlier Detection
• Metamorphic Testing of a Deep Learning based Forecaster
• A Survey on Assessing the Generalization Envelope of Deep Neural Networks: Predictive Uncertainty, Out-of-distribution and Adversarial Samples
• Deep Out-of-Distribution Uncertainty Quantification via Weight Entropy Maximization
• Reliability Quantification of Deep Reinforcement Learning-based Control
• A Multi-Scale Framework for Out-of-Distribution Detection in Dermoscopic Images
• Uncertainty Propagation in Node Classification
• Learning a Discriminant Latent Space with Neural Discriminant Analysis
• Evaluating the Use of Reconstruction Error for Novelty Localization
• Learning to Stop: Dynamic Simulation Monte-Carlo Tree Search
• Pixel Invisibility: Detecting Objects Invisible in Color Images
• DeepConsensus: using the consensus of features from multiple layers to attain robust image classification
• Distorted Representation Space Characterization Through Backpropagated Gradients
• GalilAI: Out-of-Task Distribution Detection using Causal Active Experimentation for Safe Transfer RL
• A Novel Unsupervised Post-Processing Calibration Method for DNNS with Robustness to Domain Shift
• The Robust Semantic Segmentation UNCV2023 Challenge Results
• Adversarially Robust and Explainable Model Compression with On-Device Personalization for Text Classification
• Label Smoothed Embedding Hypothesis for Out-of-Distribution Detection
• Performance Analysis of Out-of-Distribution Detection on Various Trained Neural Networks
• Probabilistic Trust Intervals for Out of Distribution Detection
• GIM: Gaussian Isolation Machines
• BED: Bi-Encoder-Based Detectors for Out-of-Distribution Detection
• OvA-INN: Continual Learning with Invertible Neural Networks
• Selective Probabilistic Classifier Based on Hypothesis Testing
• Improving robustness of classifiers by training against live traffic
• Generalized One-Class Learning Using Pairs of Complementary Classifiers
• Towards Robust Classification with Deep Generative Forests
• Toward Auto-evaluation with Confidence-based Category Relation-aware Regression
• Toward a Realistic Benchmark for Out-of-Distribution Detection
• Understanding the Role of Self-Supervised Learning in Out-of-Distribution Detection Task