Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Managing the unknown: a survey on Open Set Recognition and tangential areas

Marcos Barcina-Blanco, Jesus L. Lobo, Pablo Garcia-Bringas, Javier Del Ser

TL;DR

OSR addresses robust prediction when unknown classes appear at test time, distinguishing open-set from closed-set assumptions. It surveys discriminative and generative approaches, detailing how techniques like 1-vs-set SVM, EVT-based methods, GANs, and prototype learning tackle the open space risk $R_O(f)$ while maintaining known-class performance. The review also maps OSR to related areas (Novelty Detection, Continual Learning, OoD Detection, Uncertainty Estimation) and discusses clustering–classification hybrids, open challenges, and practical applications across vision, language, and security. The authors highlight open problems such as open space risk management, adaptive thresholds, integrated clustering–classification strategies, and incremental updating to pave the way for safer, more autonomous AI in open environments.

Abstract

In real-world scenarios classification models are often required to perform robustly when predicting samples belonging to classes that have not appeared during its training stage. Open Set Recognition addresses this issue by devising models capable of detecting unknown classes from samples arriving during the testing phase, while maintaining a good level of performance in the classification of samples belonging to known classes. This review comprehensively overviews the recent literature related to Open Set Recognition, identifying common practices, limitations, and connections of this field with other machine learning research areas, such as continual learning, out-of-distribution detection, novelty detection, and uncertainty estimation. Our work also uncovers open problems and suggests several research directions that may motivate and articulate future efforts towards more safe Artificial Intelligence methods.

Managing the unknown: a survey on Open Set Recognition and tangential areas

TL;DR

OSR addresses robust prediction when unknown classes appear at test time, distinguishing open-set from closed-set assumptions. It surveys discriminative and generative approaches, detailing how techniques like 1-vs-set SVM, EVT-based methods, GANs, and prototype learning tackle the open space risk while maintaining known-class performance. The review also maps OSR to related areas (Novelty Detection, Continual Learning, OoD Detection, Uncertainty Estimation) and discusses clustering–classification hybrids, open challenges, and practical applications across vision, language, and security. The authors highlight open problems such as open space risk management, adaptive thresholds, integrated clustering–classification strategies, and incremental updating to pave the way for safer, more autonomous AI in open environments.

Abstract

In real-world scenarios classification models are often required to perform robustly when predicting samples belonging to classes that have not appeared during its training stage. Open Set Recognition addresses this issue by devising models capable of detecting unknown classes from samples arriving during the testing phase, while maintaining a good level of performance in the classification of samples belonging to known classes. This review comprehensively overviews the recent literature related to Open Set Recognition, identifying common practices, limitations, and connections of this field with other machine learning research areas, such as continual learning, out-of-distribution detection, novelty detection, and uncertainty estimation. Our work also uncovers open problems and suggests several research directions that may motivate and articulate future efforts towards more safe Artificial Intelligence methods.
Paper Structure (20 sections, 1 equation, 1 figure, 1 table)

This paper contains 20 sections, 1 equation, 1 figure, 1 table.

Table of Contents

  1. Introduction
  2. OSR: concepts and related research areas
  3. OSR concepts and problem definition
  4. OSR in other relevant research areas
  5. Novelty Detection
  6. Continual Learning
  7. Out of Distribution Detection
  8. Uncertainty estimation
  9. OSR literature review
  10. Discriminative approaches
  11. Generative approaches
  12. Clustering-classification discriminative approaches
  13. OSR challenges and beyond
  14. Open space risk
  15. Use of thresholds
  16. ...and 5 more sections

Figures (1)

  • Figure 1: Differences between open set and open set classification models. (a) Shows the decision boundaries of an open set classifier for a problem with $N=3$ KC and $\Omega=2$ UC. All the feature space is divided between the KC, so that instances from UC arriving at inference time will be incorrectly classified, in some cases (e.g. unknown class 4) with high class probability. (b) It depicts that open set classifiers delimit the feature space that each KC occupies, allowing for the effective detection of UC.