[論文レビュー] A Comprehensive Survey of Few-shot Learning: Evolution, Applications, Challenges, and Opportunities

Yisheng Song, Ting Wang+2 more|arXiv (Cornell University)|May 13, 2022

Domain Adaptation and Few-Shot Learning被引用数 24

ひとこと要約

この論文は最近の200件超のFew-shot Learning (FSL) に関する研究を調査し、知識抽象化に基づく分類法を提案し、課題、応用、および将来の機会を詳述する。特にコンピュータビジョンにおける跨領域(クロスドメイン)およびマルチモーダルな側面に焦点を当てる。

ABSTRACT

Few-shot learning (FSL) has emerged as an effective learning method and shows great potential. Despite the recent creative works in tackling FSL tasks, learning valid information rapidly from just a few or even zero samples still remains a serious challenge. In this context, we extensively investigated 200+ latest papers on FSL published in the past three years, aiming to present a timely and comprehensive overview of the most recent advances in FSL along with impartial comparisons of the strengths and weaknesses of the existing works. For the sake of avoiding conceptual confusion, we first elaborate and compare a set of similar concepts including few-shot learning, transfer learning, and meta-learning. Furthermore, we propose a novel taxonomy to classify the existing work according to the level of abstraction of knowledge in accordance with the challenges of FSL. To enrich this survey, in each subsection we provide in-depth analysis and insightful discussion about recent advances on these topics. Moreover, taking computer vision as an example, we highlight the important application of FSL, covering various research hotspots. Finally, we conclude the survey with unique insights into the technology evolution trends together with potential future research opportunities in the hope of providing guidance to follow-up research.

研究の動機と目的

• Few-shot learning、転移学習、およびメタ学習の定義と関係を明確にし、概念的な混乱を減らす。
• 知識抽象化のレベルと課題に基づいて整理したFSLの分類法を提案する。
• 最近のFSL研究（200件以上の論文）を体系的にレビューし、進展・長所・短所を要約する。
• 跨域FSLとマルチモーダルFSLを主要な方向性として強調し、特にコンピュータビジョンにおける実用的応用と関連づける。
• FSLの将来の研究機会と進化傾向を特定する。

提案手法

• 過去3年間の200件以上のFSL論文の広範な文献調査。
• FSLにおける知識抽象化のレベルと課題に基づく新しい分類法の開発。
• アプローチを比較するための知識グラフとヒートマップの活用。
• データ拡張、転移学習、メタ学習、およびマルチモーダル学習に基づくセクション別分析。
• 画像分類、物体検出、セマンティック/インスタンス分割などのコンピュータビジョンタスクへの応用の例示。

Figure 1: The conceptual map of the survey.

実験結果

リサーチクエスチョン

• RQ1 What is few-shot learning and how does it relate to machine learning, transfer learning, and meta-learning?
• RQ2 What are the main variants and benchmarks of FSL and how have they evolved recently?
• RQ3 How can FSL be taxonomy-classified from the perspective of challenges and knowledge abstraction?
• RQ4 What are the predominant challenges and how do data augmentation, transfer learning, meta-learning, and multimodal learning address them?
• RQ5 What are the current advances and future opportunities in FSL with emphasis on cross-domain and computer vision applications?

主な発見

• The survey provides a timely, impartial overview of recent FSL advances and compares strengths and weaknesses across works.
• A novel taxonomy organizes FSL by the level of knowledge abstraction and by challenges, including a multimodal level.
• Cross-domain FSL and multimodal learning are highlighted as particularly challenging and promising directions.
• Data augmentation, transfer learning, and meta-learning are each analyzed as single-modal approaches addressing distinct facets of FSL.
• The paper documents FSL progress in computer vision tasks such as image classification, object detection, semantics segmentation, and instance segmentation.

Figure 2: The number of FSL-related papers published in prestigious journals from 2010 to the first half of 2021, excluding citations.