A Progress Report on Recent Few-Shot Learning Research

Introduction

Few-shot learning is an increasingly important research direction in machine learning. It addresses how to train high-performance models when only a very small amount of labeled data is available. In many real-world application scenarios, acquiring large-scale labeled data is often constrained by high cost, long time requirements, or privacy concerns ¹. Traditional machine learning methods usually fail to generalize effectively when data volume is insufficient ¹. Research into effective learning under few-sample conditions is therefore critical to bringing artificial intelligence into broader practical use. This report outlines the latest advances in few-shot learning published over the past year at top artificial-intelligence and machine-learning conferences (such as NeurIPS, ICML, CVPR, ICLR, and ECCV). It focuses on recent developments of key techniques — meta-learning, metric learning, pretrained models, and generative models — as well as applications of few-shot learning in computer vision, natural language processing, and related fields. It also discusses the main challenges facing current research and future research directions.

Major Advances in Few-Shot Learning Methods

Meta-learning: new methods and improvements
Meta-learning is one of the core strategies in few-shot learning. Its goal is to endow models with the ability to "learn how to learn," that is, to draw experience from a series of different tasks so they can quickly adapt to new tasks with only a few labeled samples ¹. By training on many related tasks (the meta-training phase), meta-learning enables models to acquire general knowledge or learning strategies that help them adapt and generalize rapidly when they meet new, few-sample tasks (the meta-testing phase) ⁴. Recent work shows that meta-learning remains the dominant paradigm in few-shot learning, with new methods and refinements continually emerging.
MetaFormer is a novel meta-learning framework built entirely on self-attention. It extends the Vision Transformer (ViT) so that the model can capture relations among image patches and simultaneously learn relations among samples and tasks, markedly improving few-shot image classification ²³. At its core MetaFormer proposes two new attention mechanisms: Masked Sample Attention (MSA) and Patch-grained Task Attention (PTA) ²⁴. MSA efficiently embeds sample relations into the network, uses adaptive masks to strengthen task-specific feature consistency, and flexibly switches attention across granularities. PTA maintains a dynamic knowledge pool that integrates patterns from historical tasks, capturing inter-task relations and filtering background noise ²⁴. Experiments show that MetaFormer integrates seamlessly with existing pretrained ViTs (including CLIP) and yields large gains on in-domain and cross-domain few-shot benchmarks — up to 8.77% and 6.25% over prior best methods ²³. MetaFormer's appearance shows that combining attention and Transformer architectures with meta-learning ideas is a major trend in current few-shot research.
Beyond MetaFormer, classic methods such as Model-Agnostic Meta-Learning (MAML) and Prototypical Networks remain highly important ⁶. MAML seeks a good parameter initialization so that a model can adapt to a new task with only a few gradient updates ⁷. Prototypical Networks learn a feature space in which each class is represented by a prototype; new samples are classified by their distance to the class prototypes ⁷. These methods provide a solid foundation for few-shot learning and continue to be explored and improved.
Metric-learning techniques for few-shot learning
Metric learning is another important few-shot strategy. The core idea is to learn an appropriate distance or similarity function so that, in feature space, samples of the same class lie close together and samples of different classes lie far apart, improving classification under few-shot conditions ¹. By optimizing this metric, the model better understands relations among data points and can make accurate predictions from only a few samples ¹.
Attentive Feature Regularization (AFR) is a recent metric-based few-shot method. It uses semantic information to select related classes for regularization and introduces attention at both the instance and channel levels, strengthening the representativeness and discriminability of features ³¹. Traditional manifold-regularization approaches improve performance by mixing samples from different classes, but such mixing can weaken feature representations through linear interpolation and by ignoring channel importance ³¹. AFR instead selects relevant features for regularization by computing relations among semantically different classes, and at the instance level uses attention to adaptively exploit collaborative components of related base classes, improving novel-class features. At the channel level, AFR computes channel importance weights to emphasize the key channels of discriminative patterns ³¹. Experiments show that AFR effectively raises recognition accuracy, especially in the 1-shot setting, and does so without retraining the feature extractor ³¹.
Instance-based Max-margin (IbM2) is another metric-learning method proposed for practical few-shot recognition ¹⁶. IbM2 introduces a new Practical Few-Shot Learning (pFSL) setting based on unsupervised pretrained models that can recognize many novel classes at once. Compared with traditional few-shot learning, pFSL is simpler and more effective in formulation, evaluation, and real-world use ¹⁶. To address sparse training samples, IbM2 proposes an instance-based max-margin method that applies to both the new pFSL setting and traditional FSL. Built on the Gaussian Annulus Theorem, IbM2 turns random noise applied to instances into a mechanism for achieving a max margin in multi-class pFSL (or traditional FSL) recognition ¹⁶. Experiments show that IbM2 almost always improves performance across various self-supervised pretraining methods and different few-shot or many-shot FSL tasks ¹⁶.
Using pretrained models for few-shot learning
In recent years large pretrained models — especially Vision-Language Models (VLMs) and Language Models (LMs/LLMs) — have been widely used in few-shot learning because of their strong feature representation and knowledge-transfer abilities ¹. Pretrained on massive data, they learn rich general knowledge and serve as powerful priors for few-shot learning ¹.
Low-Rank Adaptation (LoRA) is a parameter-efficient fine-tuning (PEFT) technique that adapts large pretrained models to specific tasks efficiently ⁴⁴. LoRA updates model parameters with low-rank matrices, greatly reducing computation and memory cost during fine-tuning ⁴⁴. Applying LoRA to vision-language models for few-shot learning has shown great potential ¹⁵. Compared with prompt- and adapter-based methods, LoRA yields significant gains on multiple datasets, shortens training time, and keeps hyperparameters consistent across tasks ¹⁵. Complementary Subspace Low-Rank Adaptation (Comp-LoRA) is an improvement aimed at catastrophic forgetting in few-shot VLM fine-tuning. By optimizing low-rank matrices in complementary subspaces, Comp-LoRA preserves the general vision-language alignment learned in pretraining while still learning from novel few-shot information ⁴⁷.
Prompt learning and prompt tuning are another major route to few-shot use of pretrained models, especially in natural language processing ⁴. By designing natural-language prompts, pretrained language models — particularly large language models — can be guided to complete downstream tasks with few or even zero examples, without large-scale parameter fine-tuning ⁴⁰. Adding a task description and a few examples to the input text, for instance, can effectively guide an LLM through text classification, question answering, and similar tasks ⁴⁰. Q-tuning is a prompt-tuning technique for lifelong few-shot language learning: it maintains a prompt queue and adds task-specific prompts when learning new tasks, improving transfer of knowledge from old tasks ⁴¹.
Generative models in few-shot learning
Generative models such as diffusion models and feature generators play a growing role in few-shot learning. They can synthesize additional data to expand limited training sets and improve model performance ⁴.
Diffusion models have achieved striking success in image and video generation and have also been applied to few-shot learning ⁵⁵. Few-shot diffusion models fine-tune a diffusion model pretrained on large source data using a limited set of target samples, generating samples that meet specific user needs ⁵⁵. Theoretical analysis shows that few-shot diffusion models obtain better approximation-error bounds and can reach approximate minimizers directly without complicated optimization ⁵⁵. Experiments confirm their effectiveness in image and video generation ⁵⁵.
Feature generators augment limited datasets by synthesizing new data points ⁵⁶. Some studies use class-level textual descriptions to generate visual features, addressing inaccurate embeddings caused by sparse images per class ⁵⁶. These generators are typically trained with a combination of classifier loss, discriminator loss, and a distance loss between generated features and true class embeddings, ensuring accurate same-class features and stronger overall representations ⁵⁶. Experiments show that this approach significantly raises few-shot accuracy ⁵⁶.

Recent Applications of Few-Shot Learning in Computer Vision, NLP, and Beyond

Computer vision
Few-shot learning is widely used in computer vision, especially where data are sparse or costly to obtain ¹. In medical imaging, high annotation cost and privacy concerns have driven applications in whole-slide image classification ⁵³, rare-disease diagnosis ¹, and skin-cancer detection ¹¹. In object recognition, few-shot methods let models recognize new categories from only a few examples ¹. They are also used for anomaly detection ¹ and egocentric action recognition ⁴². Recent work adapts VLMs with LoRA ¹⁵, applies IbM2 for practical few-shot recognition ¹⁶, and generates visual features from textual descriptions ⁵⁶.
Natural language processing
Few-shot learning also shows great potential in NLP, especially where large labeled corpora are unavailable ¹. Applications include sentiment analysis ¹, question answering (MinPrompt ⁷), dialogue-quality assessment ⁴¹, low-resource language processing ¹, and medical NLP ¹. Prompt learning with large language models is a leading trend in NLP few-shot research ⁴⁰. Q-tuning, for example, was proposed for lifelong few-shot language learning ⁴¹.
Other fields
Few-shot learning also has potential in robotics, helping robots learn new tasks and recognize objects from limited demonstrations ³.

Open-Source Libraries and Tools for Few-Shot Learning

Several open-source libraries and tools related to few-shot learning have been released recently, greatly advancing research and application ²⁸. The official implementation of IbM2 is available on GitHub ³⁸. CLIP-LoRA provides a convenient way to apply low-rank adaptation to CLIP for few-shot tasks, and its code is also open-sourced ⁵². Code for few-shot in-context learning with language models is available on GitHub ⁵⁰. EasyFSL is an easy-to-use and reliable few-shot learning library ⁷³. The Papers with Code platform tracks few-shot learning tasks and related code resources ²⁸. These tools make it easier for researchers and developers to run experiments and build applications.

Main Challenges and Future Directions in Few-Shot Learning

Despite significant progress, few-shot learning still faces major challenges ³. These include generalization across tasks and domains, avoiding overfitting when data are extremely scarce, handling non-IID data, bridging the semantic gap between seen and unseen classes, and designing more effective feature extraction and adaptation techniques. When pretrained models are applied to few-shot learning, avoiding catastrophic forgetting is also a major challenge ⁴⁷.

Future directions may include combining meta-learning with reinforcement learning, continual learning, and graph neural networks ⁷; exploring advances in zero-shot learning and the effective use of multimodal data ⁴; improving the interpretability of few-shot models ⁷⁵; and developing more robust and general evaluation benchmarks and datasets ⁷⁷.

New Few-Shot Learning Datasets and Benchmarks

New datasets and benchmarks have appeared to advance the field ¹⁰. Meta-Album is a large, multi-domain meta-dataset designed specifically for cross-domain few-shot research ⁷⁸. FS-Mol is a few-shot learning molecular dataset for drug discovery ⁸⁰. The Danish Fungi 2020 dataset is regarded as a challenging fine-grained few-shot classification benchmark with a large number of classes ⁸¹. These resources provide more realistic settings for evaluating generalization.

Case Studies and Practical Applications of Few-Shot Learning

Recent case studies show that few-shot learning has strong potential for real-world problems ¹. In medical diagnosis it is used for rare-disease identification and early cancer detection ¹. In robotics it enables learning of new manipulation skills from few demonstrations ³. It is applied to fraud detection ³⁰ and NLP tasks such as low-resource translation and sentiment analysis ¹. Facebook AI Research applied few-shot learning to improve face-recognition systems ⁹, and researchers at Stanford used it to accelerate drug discovery ¹¹. These cases show that few-shot learning is becoming an effective tool for problems with sparse data.

Summary of Recent Major Advances

Recent few-shot learning research has made significant progress in methods, applications, and challenge framing. On the method side, meta-learning remains the core paradigm and is being combined with advanced architectures such as Transformers, as MetaFormer shows. Metric learning has advanced through attention mechanisms and more practical settings such as IbM2. Pretrained models — especially VLMs and LLMs — play an ever larger role through parameter-efficient fine-tuning (such as LoRA) and prompt learning. Generative models, including diffusion models and feature generators, are used effectively for data augmentation to improve performance.

In applications, few-shot learning has achieved important breakthroughs in computer vision and NLP, and shows great potential in medical diagnosis, robotics, fraud detection, and other practical problems.

On the challenge side, model generalization, overfitting, and the effective use of limited data remain central concerns. Future research will likely focus on more robust and interpretable models, new learning paradigms and data modalities, and more challenging evaluation benchmarks.

In short, few-shot learning is a vigorous research field that continues to overcome the challenges of data sparsity and to show strong vitality in practical applications. As research deepens, few-shot learning is poised to play an increasingly important role in the future development of artificial intelligence.

Useful Tables

1. Overview of recent few-shot learning methods

Method	Core technique	Main idea	Venue (if applicable)	Snippet ID(s)
MetaFormer	Meta-learning / self-attention	Extends ViT to model sample and task relations jointly; proposes MSA and PTA	ICML 2024	²³
AFR	Metric learning / attention	Uses semantic info to select related classes for regularization; attention at instance and channel levels	AAAI 2024	³¹
IbM2	Metric learning / max-margin	Proposes pFSL; unsupervised pretrained models for multi-class recognition; Gaussian Annulus Theorem for instance-level max margin	CVPR 2024	¹⁶
LoRA	Pretrained models / low-rank adaptation	Efficiently fine-tunes large pretrained models (e.g. VLMs) with low-rank matrices	CVPRW 2024	¹⁵
Comp-LoRA	Pretrained models / low-rank adaptation	Optimizes low-rank matrices in complementary subspaces to avoid catastrophic forgetting	—	⁴⁷
Prompt learning / tuning	Pretrained / language models	Natural-language prompts guide pretrained LMs with few or zero samples	ACL 2024, etc.	⁴⁰
Diffusion models	Generative models	Fine-tunes pretrained diffusion models on limited target samples	NeurIPS 2024	⁵⁵
Feature generators	Generative models	Synthesizes data points (especially visual features from class-level text) to augment limited datasets	ACCV 2024	⁵⁶

2. Recent application domains of few-shot learning

Domain	Specific application	Key techniques	Snippet ID(s)
Computer vision	Medical imaging (WSI classification, rare-disease diagnosis, skin-cancer detection)	Meta-learning, metric learning, pretrained-model fine-tuning	¹
Computer vision	Object recognition	Meta-learning, metric learning, pretrained-model fine-tuning, feature generation	¹
Computer vision	Anomaly detection	Metric learning, pretrained-model fine-tuning	¹
Computer vision	Egocentric action recognition	Pretrained-model fine-tuning, multimodal learning	⁴²
NLP	Sentiment analysis	Prompt learning, pretrained-LM fine-tuning	¹
NLP	Question answering	Prompt learning, pretrained LMs	⁷
NLP	Low-resource languages	Prompt learning, pretrained-LM fine-tuning	¹
NLP	Medical NLP	Pretrained-LM fine-tuning	¹
Robotics	Object recognition and manipulation	Meta-learning, reinforcement learning	³
Finance	Fraud detection	Few-shot learning, anomaly detection	³⁰

3. Main challenges and future directions

Challenge	Description	Future direction	Snippet ID(s)
Cross-task/domain generalization	Performance drops on new tasks or domains far from training	More general meta-learning; domain adaptation	³
Overfitting	Models trained on few samples memorize data and fail to generalize	Stronger regularization; data augmentation	³
Non-IID data	Distributions shift over time or environment	Robust learning; continual learning	⁷
Semantic gap	Hard to relate novel classes to known ones	Incorporate semantic info; knowledge graphs	²²
Feature extraction & adaptation	How to extract and adapt features from pretrained models	Better feature selection and PEFT strategies	⁹
Catastrophic forgetting	Adapting to new tasks erases prior knowledge	Continual and incremental learning	⁴⁷
Model interpretability	Few-shot models are complex and hard to explain	Research on transparency and interpretability	⁸
Evaluation benchmarks	Lack of challenging benchmarks that reflect real applications	More comprehensive datasets and protocols	⁷⁷

References

Recent Advancement for Few-Shot Learning, accessed March 23, 2025, https://www.pioneerpublisher.com/jpeps/article/download/580/519
Few-shot learning based on deep learning: A survey, accessed March 23, 2025, https://www.aimspress.com/article/id/658235b8ba35de05607ed3c6
Few-Shot Learning Explained: Transforming AI With Minimal Data - Grammarly, accessed March 23, 2025, https://www.grammarly.com/blog/ai/what-is-few-shot-learning/
What is Few-Shot Learning? Strategies and Examples - Roboflow Blog, accessed March 23, 2025, https://blog.roboflow.com/few-shot-learning/
Few Shot Learning in Computer Vision: Approaches & Uses - Encord, accessed March 23, 2025, https://encord.com/blog/few-shot-learning-in-computer-vision/
Few-Shot Learning in Computer Vision: Practical Applications and Techniques, accessed March 23, 2025, https://thesciencebrigade.com/hcip/article/view/332
Few-Shot Learning: Learning with Limited Data | by Hassaan Idrees - Medium, accessed March 23, 2025, https://medium.com/@hassaanidrees7/few-shot-learning-learning-with-limited-data-9916165f5a40
What is Few-Shot Learning? Unlocking Insights with Limited Data - DataCamp, accessed March 23, 2025, https://www.datacamp.com/blog/what-is-few-shot-learning
What is Few-Shot Learning? - Lyzr AI, accessed March 23, 2025, https://www.lyzr.ai/glossaries/few-shot-learning/
Few-shot learning: everything you need to know - CUDO Compute, accessed March 23, 2025, https://www.cudocompute.com/blog/few-shot-learning-everything-you-need-to-know
Few-Shot Learning Case Studies - BytePlus, accessed March 23, 2025, https://www.byteplus.com/en/topic/400655
NeurIPS 2024 Papers, accessed March 23, 2025, https://nips.cc/virtual/2024/papers.html
ICML 2024 Conference | OpenReview, accessed March 23, 2025, https://openreview.net/group?id=ICML.cc/2024/Conference
ICML 2024 Papers, accessed March 23, 2025, https://icml.cc/virtual/2024/papers.html
Low-Rank Few-Shot Adaptation of Vision-Language Models, accessed March 23, 2025, https://openaccess.thecvf.com/content/CVPR2024W/PV/html/Zanella_Low-Rank_Few-Shot_Adaptation_of_Vision-Language_Models_CVPRW_2024_paper.html
CVPR 2024 Open Access Repository, accessed March 23, 2025, https://openaccess.thecvf.com/content/CVPR2024/html/Fu_Instance-based_Max-margin_for_Practical_Few-shot_Recognition_CVPR_2024_paper.html
Discriminative Sample-Guided and Parameter-Efficient Feature Space Adaptation for Cross-Domain Few-Shot - CVF Open Access, accessed March 23, 2025, https://openaccess.thecvf.com/content/CVPR2024/html/Perera_Discriminative_Sample-Guided_and_Parameter-Efficient_Feature_Space_Adaptation_for_Cross-Domain_Few-Shot_CVPR_2024_paper.html
CVPR 2024 Open Access Repository, accessed March 23, 2025, https://openaccess.thecvf.com/content/CVPR2024/html/Zou_Flatten_Long-Range_Loss_Landscapes_for_Cross-Domain_Few-Shot_Learning_CVPR_2024_paper.html
ICLR 2024 Papers, accessed March 23, 2025, https://iclr.cc/virtual/2024/papers.html
ICLR 2024 Neural Fine-Tuning Search for Few-Shot Learning Oral, accessed March 23, 2025, https://iclr.cc/virtual/2024/oral/19760
ECCV 2024 Papers, accessed March 23, 2025, https://eccv.ecva.net/virtual/2024/papers.html
A Comprehensive Survey of Few-shot Learning: Evolution, Applications, Challenges, and Opportunities - SciSpace, accessed March 23, 2025, https://scispace.com/papers/a-comprehensive-survey-of-few-shot-learning-evolution-2nzpufgr
One Meta-tuned Transformer is What You Need for Few-shot Learning - OpenReview, accessed March 23, 2025, https://openreview.net/pdf/06e74355729cbbba05e5627f2be61d331e35a2cb.pdf
One Meta-tuned Transformer is What You Need for Few-shot Learning, accessed March 23, 2025, https://proceedings.mlr.press/v235/yang24ah.html
Few-shot learning based on deep learning: A survey - AIMS Press, accessed March 23, 2025, https://www.aimspress.com/article/doi/10.3934/mbe.2024029?viewType=HTML
XDNet: A Few-Shot Meta-Learning Approach for Cross-Domain Visual Inspection - CVF Open Access, accessed March 23, 2025, https://openaccess.thecvf.com/content/CVPR2023W/VISION/papers/Lee_XDNet_A_Few-Shot_Meta-Learning_Approach_for_Cross-Domain_Visual_Inspection_CVPRW_2023_paper.pdf
Few-shot Learning Explained & Step-by-step How To Python Tutorial - Spot Intelligence, accessed March 23, 2025, https://spotintelligence.com/2023/06/29/few-shot-learning/
Few-Shot Learning | Papers With Code, accessed March 23, 2025, https://paperswithcode.com/task/few-shot-learning
Attentive Pooling Network for Few-Shot Learning - JMIS, accessed March 23, 2025, https://www.jmis.org/archive/view_article?pid=jmis-9-4-269
Few-shot classification case studies: revolutionizing machine learning with limited data - BytePlus, accessed March 23, 2025, https://www.byteplus.com/en/topic/400710
Boosting Few-Shot Learning via Attentive Feature Regularization, accessed March 23, 2025, https://ojs.aaai.org/index.php/AAAI/article/view/28614/29193
Boosting Few-Shot Learning via Attentive Feature Regularization - arXiv, accessed March 23, 2025, https://arxiv.org/html/2403.17025v1
[2403.17025] Boosting Few-Shot Learning via Attentive Feature Regularization - arXiv, accessed March 23, 2025, https://arxiv.org/abs/2403.17025
Boosting Few-Shot Learning via Attentive Feature Regularization - ResearchGate, accessed March 23, 2025, https://www.researchgate.net/publication/379280333_Boosting_Few-Shot_Learning_via_Attentive_Feature_Regularization
Instance-based Max-margin for Practical Few-shot Recognition - CVF Open Access, accessed March 23, 2025, https://openaccess.thecvf.com/content/CVPR2024/papers/Fu_Instance-based_Max-margin_for_Practical_Few-shot_Recognition_CVPR_2024_paper.pdf
[2305.17368] Instance-based Max-margin for Practical Few-shot Recognition - arXiv, accessed March 23, 2025, https://arxiv.org/abs/2305.17368
Instance-based Max-margin for Practical Few-shot Recognition - IEEE Computer Society, accessed March 23, 2025, https://www.computer.org/csdl/proceedings-article/cvpr/2024/530000ad674/20hTzGdojDO
heekhero/IbM2: This repo is the official implementation of ... - GitHub, accessed March 23, 2025, https://github.com/heekhero/IbM2
CVPR Poster Instance-based Max-margin for Practical Few-shot Recognition, accessed March 23, 2025, https://cvpr.thecvf.com/virtual/2024/poster/29970
NeurIPS Poster The Representation Landscape of Few-Shot ..., accessed March 23, 2025, https://neurips.cc/virtual/2024/poster/93666
Few-shot learning - Amazon Science, accessed March 23, 2025, https://www.amazon.science/tag/few-shot-learning
Multimodal Cross-Domain Few-Shot Learning for Egocentric Action Recognition - arXiv, accessed March 23, 2025, https://arxiv.org/abs/2405.19917
[2301.02419] Exploring Efficient Few-shot Adaptation for Vision Transformers - arXiv, accessed March 23, 2025, https://arxiv.org/abs/2301.02419
NORA: Nested Low-Rank Adaptation for Efficient Fine-Tuning Large Models - OpenReview, accessed March 23, 2025, https://openreview.net/pdf/fbf030c71d4228a1a08a5f0b9eb589a08bf974ae.pdf
Low-Rank Few-Shot Adaptation of Vision-Language Models | AI Research Paper Details, accessed March 23, 2025, https://www.aimodels.fyi/papers/arxiv/low-rank-few-shot-adaptation-vision-language
[PDF] Low-Rank Few-Shot Adaptation of Vision-Language Models | Semantic Scholar, accessed March 23, 2025, https://www.semanticscholar.org/paper/Low-Rank-Few-Shot-Adaptation-of-Vision-Language-Zanella-Ayed/42012c4e9dc8d787adb101548cefc42b8a4507d9
Complementary Subspace Low-Rank Adaptation of Vision-Language Models for Few-Shot Classification - ResearchGate, accessed March 23, 2025, https://www.researchgate.net/publication/388422577_Complementary_Subspace_Low-Rank_Adaptation_of_Vision-Language_Models_for_Few-Shot_Classification
Low-Rank Few-Shot Adaptation of Vision-Language Models - IEEE Computer Society, accessed March 23, 2025, https://www.computer.org/csdl/proceedings-article/cvprw/2024/654700b593/20A2rXUClFe
Few-shot Learning in Natural Language Processing - NTT Data, accessed March 23, 2025, https://www.nttdata.com/global/en/-/media/nttdataglobal/1_files/services/data-and-intelligence/few-shot_learning_in_natural_language_processing.pdf?rev=37e2bca167ad474485aa5de0978851f3
Few-shot Learning With Language Models - GitHub, accessed March 23, 2025, https://github.com/tonyzhaozh/few-shot-learning
Exploring Zero-Shot and Few-Shot Learning in Generative AI - [x]cube LABS, accessed March 23, 2025, https://www.xcubelabs.com/blog/exploring-zero-shot-and-few-shot-learning-in-generative-ai/
MaxZanella/CLIP-LoRA: An easy way to apply LoRA to ... - GitHub, accessed March 23, 2025, https://github.com/MaxZanella/CLIP-LoRA
NeurIPS Poster FAST: A Dual-tier Few-Shot Learning Paradigm for ..., accessed March 23, 2025, https://neurips.cc/virtual/2024/poster/96289
LLM Learning Corner: Zero-Shot & Few-Shot Learning - Devolved AI, accessed March 23, 2025, https://www.devolvedai.com/posts/06-zero-shot-few-shot-learning
NeurIPS Poster Few-Shot Diffusion Models Escape the Curse of ..., accessed March 23, 2025, https://neurips.cc/virtual/2024/poster/95694
Feature Generator for Few-Shot Learning - CVF Open Access, accessed March 23, 2025, https://openaccess.thecvf.com/content/ACCV2024/papers/Kanagalingam_Feature_Generator_for_Few-Shot_Learning_ACCV_2024_paper.pdf
[2409.14141] A Feature Generator for Few-Shot Learning - arXiv, accessed March 23, 2025, https://arxiv.org/abs/2409.14141
A Feature Generator for Few-Shot Learning - arXiv, accessed March 23, 2025, https://arxiv.org/html/2409.14141
A Feature Generator for Few-Shot Learning - ResearchGate, accessed March 23, 2025, https://www.researchgate.net/publication/384267222_A_Feature_Generator_for_Few-Shot_Learning
[PDF] A Feature Generator for Few-Shot Learning - Semantic Scholar, accessed March 23, 2025, https://www.semanticscholar.org/paper/A-Feature-Generator-for-Few-Shot-Learning-Kanagalingam-Pathmanathan/c846e73707bc4db66bb436bc6cdb45f19088c249
[ECCV 2024] Multimodal Cross-Domain Few-Shot Learning for ..., accessed March 23, 2025, https://www.youtube.com/watch?v=f4ZK66xI3LU
Masashi Hatano, accessed March 23, 2025, https://masashi-hatano.github.io/
milvus.io, accessed March 23, 2025, https://milvus.io/ai-quick-reference/how-can-fewshot-learning-be-applied-in-computer-vision
How can few-shot learning be applied in computer vision? - Milvus, accessed March 23, 2025, https://milvus.io/ai-quick-reference/how-can-fewshot-learning-be-applied-in-computer-vision
Recent Advances in Few-Shot Learning | Restackio, accessed March 23, 2025, https://www.restack.io/p/few-shot-learning-answer-advances-applications-cat-ai
Few-Shot Learning for Clinical Natural Language Processing Using Siamese Neural Networks: Algorithm Development and Validation Study, accessed March 23, 2025, https://pmc.ncbi.nlm.nih.gov/articles/PMC11041484/
Learning to explain is a good biomedical few-shot learner | Bioinformatics | Oxford Academic, accessed March 23, 2025, https://academic.oup.com/bioinformatics/article/40/10/btae589/7808853
What are the typical applications of few-shot learning? - Milvus, accessed March 23, 2025, https://milvus.io/ai-quick-reference/what-are-the-typical-applications-of-fewshot-learning
Multimodal Cross-Domain Few-Shot Learning for Egocentric Action Recognition - arXiv, accessed March 23, 2025, https://arxiv.org/html/2405.19917v1
Multimodal Cross-Domain Few-Shot Learning for Egocentric Action Recognition, accessed March 23, 2025, https://www.researchgate.net/publication/381005611_Multimodal_Cross-Domain_Few-Shot_Learning_for_Egocentric_Action_Recognition
Multimodal Cross-Domain Few-Shot Learning for Egocentric Action Recognition, accessed March 23, 2025, https://www.ecva.net/papers/eccv_2024/papers_ECCV/papers/04830.pdf
Everything you need to know about Few-Shot Learning | DigitalOcean, accessed March 23, 2025, https://www.digitalocean.com/community/tutorials/few-shot-learning
EasyFSL: Easy and Reliable Code for Few-Shot Learning - Theodo Data & AI, accessed March 23, 2025, https://data-ai.theodo.com/en/technical-blog/easyfsl-code-few-shot-learning
How can few-shot learning be used to identify new diseases in healthcare? - Milvus Blog, accessed March 23, 2025, https://blog.milvus.io/ai-quick-reference/how-can-fewshot-learning-be-used-to-identify-new-diseases-in-healthcare
What are the challenges of using few-shot learning in real-world applications? - Milvus, accessed March 23, 2025, https://milvus.io/ai-quick-reference/what-are-the-challenges-of-using-fewshot-learning-in-realworld-applications
[2402.01440] A Survey of Few-Shot Learning on Graphs: from Meta-Learning to Pre-Training and Prompt Learning - arXiv, accessed March 23, 2025, https://arxiv.org/abs/2402.01440
A Survey of Few-Shot Learning for Biomedical Time Series | AI Research Paper Details, accessed March 23, 2025, https://www.aimodels.fyi/papers/arxiv/survey-few-shot-learning-biomedical-time-series
Meta-Album: Multi-domain Meta-Dataset for Few-Shot Image Classification - NeurIPS, accessed March 23, 2025, https://proceedings.neurips.cc/paper_files/paper/2022/file/1585da86b5a3c4fb15520a2b3682051f-Paper-Datasets_and_Benchmarks.pdf
Announcing Meta-Dataset: A Dataset of Datasets for Few-Shot Learning - Google Research, accessed March 23, 2025, https://research.google/blog/announcing-meta-dataset-a-dataset-of-datasets-for-few-shot-learning/
FS-Mol: A Few-Shot Learning Dataset of Molecules, accessed March 23, 2025, https://datasets-benchmarks-proceedings.neurips.cc/paper/2021/hash/8d3bba7425e7c98c50f52ca1b52d3735-Abstract-round2.html
Few-Shot Image Classification Benchmarks Are Too Far From Reality: Build Back Better With Semantic Task Sampling, accessed March 23, 2025, https://openaccess.thecvf.com/content/CVPR2022W/VDU/papers/Bennequin_Few-Shot_Image_Classification_Benchmarks_Are_Too_Far_From_Reality_Build_CVPRW_2022_paper.pdf