A Comprehensive Survey of Few-shot Learning: Evolution, Applications, Challenges, and Opportunities

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 remains a serious challenge. In this context, we extensively investigated 200+ FSL papers published in top journals and conferences in the past three years, aiming to present a timely and comprehensive overview of the most recent advances in FSL with a fresh perspective, and to provide an impartial comparison of the strengths and weaknesses of existing work. To avoid conceptual confusion, we first elaborate and contrast a set of relevant concepts including few-shot learning, transfer learning, and meta-learning. Then, we inventively extract prior knowledge related to few-shot learning in the form of a pyramid, which summarizes and classifies previous work in detail from the perspective of challenges. Furthermore, to enrich this survey, we present in-depth analysis and insightful discussions of recent advances in each subsection. What’s more, 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 technology trends and potential future research opportunities to guide FSL follow-up research.

Learning Expressive Prompting With Residuals for Vision Transformers

“…Few-Shot Classification: Few-shot classification has re-ceived a lot of attention in recent years. While a variety of approaches [75] have been proposed, the most successful ones seek to transfer positive knowledge either by finetuning [1,12,79] or meta-learning [17,19,21,24,38,62,74,77,81]. Finetuning based few-shot learners can be viewed as specialists that perform well on the target domain [9,27,79], but suffer from catastrophic forgetting [70] on the base domain.…”

Deep Learning for Cross-Domain Few-Shot Visual Recognition: A Survey

“…The results of other methods within this category are even lower. When comparing the use of ResNet10 (90 25.25% of ChestX) as the backbone for [124] on BSCD-FSL, it is observed that while increasing the depth of the network enhances performance on near-domain datasets (CropDiseases, EuroSAT), it deteriorates performance on distant-domain datasets (ISIC, ChestX). As such, the best balance of near-domain and distant-domain performance is achieved when using ResNet10 as the backbone.…”

“…Therefore, the goal of FSL is to leverage prior knowledge to learn new tasks with only a few labeled samples, which has attracted significant attention due to its crucial industrial and academic applications. Since the introduction of this problem in 2006 [18], numerous research methods have been proposed [61,75,87,90,114].…”

Effective Adaptation into New Environment with Few Shots: Applications to OFDM Receiver Design

“…FSL is an emerging paradigm that addresses these challenges by producing creative work on data, models, and algorithms. According to [8], FSL techniques can be divided into data augmentation, multimodal learning, meta-learning [9]- [11], and transfer learning (TL) [12]- [14]. One critical issue of FSL in wireless communications is that the data from different environments are heterogeneous and the samples of a new environment are small.…”