Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Few-Shot Learning in the Open World: A Review and Beyond

Hui Xue, Yuexuan An, Yongchun Qin, Wenqian Li, Yixin Wu, Yongjuan Che, Pengfei Fang, Minling Zhang

TL;DR

Open-world Few-Shot Learning (OFSL) extends traditional FSL to dynamic, uncertain environments by handling unseen concepts across three variants: varying instances, varying classes, and varying distributions. The paper provides a unified taxonomy and surveys representative methods across NFSL, FSOSR, IFSL, and CDFSL, emphasizing standardized benchmarks and cross-domain comparisons. It analyzes data augmentation, parameter optimization, and feature processing strategies that enable robust generalization with limited labeled data. The work highlights open challenges—such as multi-label concepts, adversarial robustness, and imbalanced data—and outlines future research directions to advance practical OFSL systems in real-world applications.

Abstract

Human intelligence is characterized by our ability to absorb and apply knowledge from the world around us, especially in rapidly acquiring new concepts from minimal examples, underpinned by prior knowledge. Few-shot learning (FSL) aims to mimic this capacity by enabling significant generalizations and transferability. However, traditional FSL frameworks often rely on assumptions of clean, complete, and static data, conditions that are seldom met in real-world environments. Such assumptions falter in the inherently uncertain, incomplete, and dynamic contexts of the open world. This paper presents a comprehensive review of recent advancements designed to adapt FSL for use in open-world settings. We categorize existing methods into three distinct types of open-world few-shot learning: those involving varying instances, varying classes, and varying distributions. Each category is discussed in terms of its specific challenges and methods, as well as its strengths and weaknesses. We standardize experimental settings and metric benchmarks across scenarios, and provide a comparative analysis of the performance of various methods. In conclusion, we outline potential future research directions for this evolving field. It is our hope that this review will catalyze further development of effective solutions to these complex challenges, thereby advancing the field of artificial intelligence.

Towards Few-Shot Learning in the Open World: A Review and Beyond

TL;DR

Open-world Few-Shot Learning (OFSL) extends traditional FSL to dynamic, uncertain environments by handling unseen concepts across three variants: varying instances, varying classes, and varying distributions. The paper provides a unified taxonomy and surveys representative methods across NFSL, FSOSR, IFSL, and CDFSL, emphasizing standardized benchmarks and cross-domain comparisons. It analyzes data augmentation, parameter optimization, and feature processing strategies that enable robust generalization with limited labeled data. The work highlights open challenges—such as multi-label concepts, adversarial robustness, and imbalanced data—and outlines future research directions to advance practical OFSL systems in real-world applications.

Abstract

Human intelligence is characterized by our ability to absorb and apply knowledge from the world around us, especially in rapidly acquiring new concepts from minimal examples, underpinned by prior knowledge. Few-shot learning (FSL) aims to mimic this capacity by enabling significant generalizations and transferability. However, traditional FSL frameworks often rely on assumptions of clean, complete, and static data, conditions that are seldom met in real-world environments. Such assumptions falter in the inherently uncertain, incomplete, and dynamic contexts of the open world. This paper presents a comprehensive review of recent advancements designed to adapt FSL for use in open-world settings. We categorize existing methods into three distinct types of open-world few-shot learning: those involving varying instances, varying classes, and varying distributions. Each category is discussed in terms of its specific challenges and methods, as well as its strengths and weaknesses. We standardize experimental settings and metric benchmarks across scenarios, and provide a comparative analysis of the performance of various methods. In conclusion, we outline potential future research directions for this evolving field. It is our hope that this review will catalyze further development of effective solutions to these complex challenges, thereby advancing the field of artificial intelligence.
Paper Structure (40 sections, 2 equations, 9 figures, 4 tables)

This paper contains 40 sections, 2 equations, 9 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Problem Definition
  3. Few-shot Learning
  4. Open-world Few-Shot Learning
  5. Few-Shot Learning with Varying Instances in the Open World
  6. Few-Shot Learning with Varying Classes in the Open World
  7. Few-Shot Learning with Varying Distributions in the Open World
  8. Few-Shot Leaning with Varying Instances in the Open World: Noisy Few-Shot Learning
  9. Parameter Optimization
  10. Sample Selection
  11. Few-Shot Leaning with Varying Instances in the Open World: Few-Shot Open-Set Recognition
  12. Data Augmentation
  13. Evaluation Metric
  14. Feature Processing.
  15. Few-Shot Learning with Varying Classes in the Open World: Incremental Few-Shot Learning
  16. ...and 25 more sections

Figures (9)

  • Figure 1: Chronological milestones on FSL and OFSL, including representative FSL and OFSL approaches and the related benchmarks. FSL was first noticed as a topic in 2003 LiF2003. Subsequently, it underwent continuous developments SiameseNetCloser. After that, different open-world few-shot learning scenarios have been proposed successively, including few-shot learning with varying instances PEELERRFSLTraNFLPCLIDEAL, few-shot learning with varying classes TOPICWillesJ2024 and few-shot learning with varying distributions c76_triantafillou2019metac75_guo2020broaderc28_tseng2020cross in the open world.
  • Figure 2: The structural relationships of the existing two types of methods and their subclasses, as well as the difficulties and challenges faced by NFSL problems.
  • Figure 3: An overall diagram of existing NFSL researches. (a) Overview of the NFSL setting.(b) Parameter Optimization methods are mainly divided into Gradient-based and Metric-based methods. (c)The sample selection methods are mainly divided into Test-Time Adaptation and Meta-Training Weighting.
  • Figure 4: The structural relationships of the existing three types of methods and their subclasses, as well as the difficulties and challenges faced by FSOSR problems.
  • Figure 5: An overall diagram of existing FSOSR researches. (a) Overview of the FSOSR setting.(b) Data Augmentation methods are mainly divided into Generative Model and Auxiliary Information. (c) Metric Evaluation methods include Distance Modification and Density Distribution. (d)Feature Processing methods compare the differences between the reconstructed sets before and after using transformation functions.
  • ...and 4 more figures