Reducing Network Agnostophobia

Abstract

Agnostophobia, the fear of the unknown, can be experienced by deep learning engineers while applying their networks to real-world applications. Unfortunately, network behavior is not well defined for inputs far from a networks training set. In an uncontrolled environment, networks face many instances that are not of interest to them and have to be rejected in order to avoid a false positive. This problem has previously been tackled by researchers by either a) thresholding softmax, which by construction cannot return "none of the known classes", or b) using an additional background or garbage class. In this paper, we show that both of these approaches help, but are generally insufficient when previously unseen classes are encountered. We also introduce a new evaluation metric that focuses on comparing the performance of multiple approaches in scenarios where such unseen classes or unknowns are encountered. Our major contributions are simple yet effective Entropic Open-Set and Objectosphere losses that train networks using negative samples from some classes. These novel losses are designed to maximize entropy for unknown inputs while increasing separation in deep feature space by modifying magnitudes of known and unknown samples. Experiments on networks trained to classify classes from MNIST and CIFAR-10 show that our novel loss functions are significantly better at dealing with unknown inputs from datasets such as Devanagari, NotMNIST, CIFAR-100, and SVHN.

Neural Information Processing Systems (NeurIPS) 2018

Cited by in corpus (37)

• Recent Advances in Open Set Recognition: A Survey
• Adversarial Reciprocal Points Learning for Open Set Recognition
• Learning Open Set Network with Discriminative Reciprocal Points
• Contrastive Out-of-Distribution Detection for Pretrained Transformers
• SSD: A Unified Framework for Self-Supervised Outlier Detection
• A Survey on Open Set Recognition
• Open-Set Recognition: a Good Closed-Set Classifier is All You Need?
• Unified Probabilistic Deep Continual Learning through Generative Replay and Open Set Recognition
• Generalized ODIN: Detecting Out-of-distribution Image without Learning from Out-of-distribution Data
• Entropic Out-of-Distribution Detection: Seamless Detection of Unknown Examples
• Incremental Deep Neural Network Learning using Classification Confidence Thresholding
• A Unified Survey on Anomaly, Novelty, Open-Set, and Out-of-Distribution Detection: Solutions and Future Challenges
• Automatic Open-World Reliability Assessment
• Density of States Estimation for Out-of-Distribution Detection
• Self-Supervised Features Improve Open-World Learning
• Open Set Medical Diagnosis
• Controlling Risk of Web Question Answering
• Revisiting the Evaluation of Uncertainty Estimation and Its Application to Explore Model Complexity-Uncertainty Trade-Off
• Stationary Activations for Uncertainty Calibration in Deep Learning
• A Critical Evaluation of Open-World Machine Learning
• Better the Devil you Know: An Analysis of Evasion Attacks using Out-of-Distribution Adversarial Examples
• Few-shot Open-set Recognition by Transformation Consistency
• Class Anchor Clustering: a Loss for Distance-based Open Set Recognition
• Bayesian OOD detection with aleatoric uncertainty and outlier exposure
• Enhanced Isotropy Maximization Loss: Seamless and High-Performance Out-of-Distribution Detection Simply Replacing the SoftMax Loss
• Improved Robustness to Open Set Inputs via Tempered Mixup
• Knowing the No-match: Entity Alignment with Dangling Cases
• DeepStreamCE: A Streaming Approach to Concept Evolution Detection in Deep Neural Networks
• Annotation-Efficient Untrimmed Video Action Recognition
• Making Table Understanding Work in Practice
• MMF: A loss extension for feature learning in open set recognition
• One-Class Meta-Learning: Towards Generalizable Few-Shot Open-Set Classification
• A novel network training approach for open set image recognition
• Visual and Semantic Prototypes-Jointly Guided CNN for Generalized Zero-shot Learning
• A Review of Computer Vision Methods in Network Security
• Joint Distribution across Representation Space for Out-of-Distribution Detection
• Deep Learning and Open Set Malware Classification: A Survey