Few-shot Learning Research Articles

Exploring the Gap: The Challenge of Achieving Human-like Generalization for Concept-based Translation Instruction Using Large Language Models

Our study utilizes concept description instructions and few-shot learning examples to examine the effectiveness of a large language model (GPT-4) in generating Chinese-to-English translations that embody related translation concepts. We discovered that human language experts possess superior abductive reasoning skills compared to GPT-4. Therefore, it is crucial for humans to employ abductive reasoning to craft more detailed instructions and infuse additional logic into exemplary prompts, a step essential for guiding a large language model effectively, in contrast to the more intuitive understanding a human expert might have. This approach would make the prompt engineering process more complicated and less human-like. Emphasizing domain-specific abductive reasoning stands out as a crucial aspect of human-like learning that AI/ML systems based on large language models should aim to replicate.

Boosting few-shot rare skin disease classification via self-supervision and distribution calibration.

Due to the difficulty in obtaining clinical samples and the high cost of labeling, rare skin diseases are characterized by data scarcity, making training deep neural networks for classification challenging. In recent years, few-shot learning has emerged as a promising solution, enabling models to recognize unseen disease classes by limited labeled samples. However, most existing methods ignored the fine-grained nature of rare skin diseases, resulting in poor performance when generalizing to highly similar classes. Moreover, the distributions learned from limited labeled data are biased, severely impairing the model's generalizability. This paper proposes a self-supervision distribution calibration network (SS-DCN) to address the above issues. Specifically, SS-DCN adopts a multi-task learning framework during pre-training. By introducing self-supervised tasks to aid in supervised learning, the model can learn more discriminative and transferable visual representations. Furthermore, SS-DCN applied an enhanced distribution calibration (EDC) strategy, which utilizes the statistics of base classes with sufficient samples to calibrate the bias distribution of novel classes with few-shot samples. By generating more samples from the calibrated distribution, EDC can provide sufficient supervision for subsequent classifier training. The proposed method is evaluated on three public skin disease datasets(i.e., ISIC2018, Derm7pt, and SD198), achieving significant performance improvements over state-of-the-art methods.

Building Trustworthy Generative Artificial Intelligence for Diabetes Care and Limb Preservation: A Medical Knowledge Extraction Case.

Large language models (LLMs) offer significant potential in medical information extraction but carry risks of generating incorrect information. This study aims to develop and validate a retriever-augmented generation (RAG) model that provides accurate medical knowledge about diabetes and diabetic foot care to laypersons with an eighth-grade literacy level. Improving health literacy through patient education is paramount to addressing the problem of limb loss in the diabetic population. In addition to affecting patient well-being through improved outcomes, improved physician well-being is an important outcome of a self-management model for patient health education. We used an RAG architecture and built a question-and-answer artificial intelligence (AI) model to extract knowledge in response to questions pertaining to diabetes and diabetic foot care. We utilized GPT-4 by OpenAI, with Pinecone as a vector database. The NIH National Standards for Diabetes Self-Management Education served as the basis for our knowledge base. The model's outputs were validated through expert review against established guidelines and literature. Fifty-eight keywords were used to select 295 articles and the model was tested against 175 questions across topics. The study demonstrated that with appropriate content volume and few-shot learning prompts, the RAG model achieved 98% accuracy, confirming its capability to offer user-friendly and comprehensible medical information. The RAG model represents a promising tool for delivering reliable medical knowledge to the public which can be used for self-education and self-management for diabetes, highlighting the importance of content validation and innovative prompt engineering in AI applications.

Zero-shot knowledge transfer for seismic damage diagnosis through multi-channel 1D CNN integrated with autoencoder-based domain adaptation

Accurate and timely structural damage diagnosis is crucial to efficient disaster response and city renovation in post-earthquake events. The scarcity of labeled data hinders the powerful deep learning techniques from in-domain damage detection on target structures. Cross-domain transfer learning has emerged as a captivating strategy through digging knowledge from the abundant source domain to detect the damage in the target domain. However, the heterogeneity among multi-domain structures poses the challenge in seismic damage diagnosis. This study proposes a novel zero-shot knowledge transfer approach for seismic damage diagnosis through multi-channel one-dimensional convolutional neural networks (1D CNN) integrated with deep autoencoder (DAE)-based domain adaptation (DA). The framework consists of three modules, namely, data preprocessor adaptive to seismic vibration signals, DAE-based DA module, and damage diagnosis via multi-channel 1D CNN. The DA module is customized to seamlessly translate the unseen target-domain data to mimic latent representation via a DAE pretrained on the source data, thus realizing rigorous zero-shot learning. Imbalanced data distribution is also considered during the network training and testing. Two representative phases of knowledge transfer are performed to substantiate the knowledge transferability of the proposed method. The first phase involves multi-class damage quantification on two ASCE benchmark models from the simplified model to the refined one, and the second phase conducts binary damage detection on a three-story reinforced frame structure from the finite element numerical model to the laboratory-tested physical model. Both examples show that the proposed method exhibits high prediction accuracy and a lower false negative rate in achieving zero-shot knowledge transfer for cross-domain structural damage diagnosis. With a delicate network design for diverse data, the proposed knowledge transfer framework can be further extended from the present zero-shot approach to the few-shot learning paradigm, thus suggesting a feasible algorithm adaptability and promising engineering applicability.

Few-shot biomedical relation extraction using data augmentation and domain information

Relation extraction (RE) plays a pivotal role in biomedical information extraction. However, traditional approaches are often limited by high data annotation costs and extensive time investments. To address this challenge, this study proposes an innovative few-shot learning approach for biomedical RE tasks that utilizes data augmentation and domain information. Specifically, this method enhances the diversity of training data through a synthesized data augmentation strategy. At the same time, it combines domain information with prompt learning techniques. By incorporating domain information, the model’s generalization capability for rare data instances is improved. Furthermore, the prompt learning approach, by integrating prompt templates into the model input, guides the pre-trained language model to more accurately adapt to the RE task. Our model demonstrated exceptional performance in three different biomedical RE tasks, particularly on the I2B2-2010 RE dataset. In the 1-shot, 10-shot, and 50-shot settings, the F1 scores increased by 14.38%, 9.50%, and 6.05%, respectively.

A Novel Fault Diagnosis Method of High-Speed Train Based on Few-Shot Learning.

Ensuring the safe and stable operation of high-speed trains necessitates real-time monitoring and diagnostics of their suspension systems. While machine learning technology is widely employed for industrial equipment fault diagnosis, its effective application relies on the availability of a large dataset with annotated fault data for model training. However, in practice, the availability of informational data samples is often insufficient, with most of them being unlabeled. The challenge arises when traditional machine learning methods encounter a scarcity of training data, leading to overfitting due to limited information. To address this issue, this paper proposes a novel few-shot learning method for high-speed train fault diagnosis, incorporating sensor-perturbation injection and meta-confidence learning to improve detection accuracy. Experimental results demonstrate the superior performance of the proposed method, which introduces perturbations, compared to existing methods. The impact of perturbation effects and class numbers on fault detection is analyzed, confirming the effectiveness of our learning strategy.

Data-efficient multimodal human action recognition for proactive human–robot collaborative assembly: A cross-domain few-shot learning approach

With the recent vision of Industry 5.0, the cognitive capability of robots plays a crucial role in advancing proactive human–robot collaborative assembly. As a basis of the mutual empathy, the understanding of a human operator’s intention has been primarily studied through the technique of human action recognition. Existing deep learning-based methods demonstrate remarkable efficacy in handling information-rich data such as physiological measurements and videos, where the latter category represents a more natural perception input. However, deploying these methods in new unseen assembly scenarios requires first collecting abundant case-specific data. This leads to significant manual effort and poor flexibility. To deal with the issue, this paper proposes a novel cross-domain few-shot learning method for data-efficient multimodal human action recognition. A hierarchical data fusion mechanism is designed to jointly leverage the skeletons, RGB images and depth maps with complementary information. Then a temporal CrossTransformer is developed to enable the action recognition with very limited amount of data. Lightweight domain adapters are integrated to further improve the generalization with fast finetuning. Extensive experiments on a real car engine assembly case show the superior performance of proposed method over state-of-the-art regarding both accuracy and finetuning efficiency. Real-time demonstrations and ablation study further indicate the potential of early recognition, which is beneficial for the robot procedures generation in practical applications. In summary, this paper contributes to the rarely explored realm of data-efficient human action recognition for proactive human–robot collaboration.

Towards intrusion detection in IoT using Few-shot learning

The Internet of Things (IoT) is an emerging technology that covers various domains and has become an essential part of the upcoming technological revolution. IoT applications include healthcare, smart-cities, smart-cars, industries, quality of life, and several other fields. IoT typically consists of lightweight sensor devices that facilitate procedures such as automation, real-time trackable data collection, and data-driven decisions. However, securing IoT networks is an accessible research area for several reasons. The main security challenges are limited resources that are incapable of dealing with complex and advanced security tools; and lack of required data for training the security systems like Intrusion detection systems as a result of their heterogeneous nature. This research proposed a Few-shot learning IoT intrusion detection system model based on a Siamese network to overcome the above limitation. The model aims to classify and distinguish normal and attacked traffic. The experiment utilized an IoT dataset in different scenarios to analyze and validate the behavior with three categories with different numbers of data in each. The performance result achieves more than 99% accuracy and shows an efficient detection ability using only less than 1% of the dataset.

Supervised and Few-Shot Learning for Aspect-Based Sentiment Analysis of Instruction Prompt

Supervised and Few-Shot Learning for Aspect-Based Sentiment Analysis of Instruction Prompt

Which images to label for few-shot medical image analysis?

The success of deep learning methodologies hinges upon the availability of meticulously labeled extensive datasets. However, when dealing with medical images, the annotation process for such abundant training data often necessitates the involvement of experienced radiologists, thereby consuming their limited time resources. In order to alleviate this burden, few-shot learning approaches have been developed, which manage to achieve competitive performance levels with only several labeled images. Nevertheless, a crucial yet previously overlooked problem in few-shot learning is about the selection of template images for annotation before learning, which affects the final performance. In this study, we propose a novel TEmplate Choosing Policy (TECP) that aims to identify and select “the most worthy” images for annotation, particularly within the context of multiple few-shot medical tasks, including landmark detection, anatomy detection, and anatomy segmentation. TECP is composed of four integral components: (1) Self-supervised training, which entails training a pre-existing deep model to extract salient features from radiological images; (2) Alternative proposals for localizing informative regions within the images; and (3) Representative Score Estimation, which involves the evaluation and identification of the most representative samples or templates. (4) Ranking, which rank all candidates and select one with highest representative score. The efficacy of the TECP approach is demonstrated through a series of comprehensive experiments conducted on multiple public datasets. Across all three medical tasks, the utilization of TECP yields noticeable improvements in model performance.

Rapid in-situ calibration of computational micro-spectrometer with few-shot meta-learning

Computational micro-spectrometers comprised of detector arrays and encoding structure arrays, such as on-chip Fabry-Perot (FP) cavity filters, have great potential in many in-situ applications owing to their compact size and snapshot imaging ability. Given manufacturing deviation and environmental influence are inevitable, easy and effective calibration for spectrometer is necessary, especially for in-situ applications. Currently calibration strategies based on iterative algorithms or neural networks require accurate measurements of pixel-level (spectral) encoding functions through monochromator or large amounts of standard samples. These procedures are time-consuming and expensive, thereby impeding in-situ applications. Meta-learning algorithms with few-shot learning ability can address this challenge by incorporating the prior knowledge in the simulated dataset. In this work, we propose a meta-learning algorithm free of measuring encoding function or large amounts of standard samples to calibrate a micro-spectrometer with manufacturing deviation effectively. Our micro-spectrometer comprises 16 types of FP filters covering a wavelength range of 550-720 nm. The center wavelength of each filter type deviates from the design up to 6 nm. After calibration with 15 different color data, the average reconstruction error on the test dataset decreased from 7.2 × 10 − 3 to 1.2 × 10 − 3, and further decreased to 9.4 × 10 − 4 when the calibration data increased to 24. The performance is comparable to algorithms trained with measured encoding function both in reconstruction error and generalization ability. We estimated that the cost of in-situ calibration through reflectance measurements of color chart decreased to one percent of the cost through monochromator measurements. By exploiting prior deviation information in simulation data with meta-learning, the efficiency and cost of calibration are significantly improved, thereby facilitating the large-scale production and in-situ application of micro-spectrometers.

NERO: NEural algorithmic reasoning for zeRO-day attack detection in the IoT: A hybrid approach

Anomaly detection approaches for network intrusion detection learn to identify deviations from normal behavior on a data-driven basis. However, current approaches strive to infer the degree of abnormality of out-of-distribution samples when these appertain to different zero-day attacks. Inspired by the successes of the neural algorithmic reasoning paradigm to leverage the generalization of rule-based behavior, this paper presents a deep learning strategy for solving zero-day network attack detection and categorization. Moreover, focusing on the particular scenario of the Internet of Things (IoT), the privacy preservation requirement may imply a low training data regime for any learning algorithm. To this respect, the presented framework uses metric-based meta-learning to achieve few-shot learning capabilities. The presented pipeline is called NERO, as it imports the encode-process-decode architecture from the NEural algorithmic reasoning blueprint to converge zeRO-day attack detection policies within constrained training data.

Local descriptor-based spatial cross attention network for few-shot learning

Local descriptor-based spatial cross attention network for few-shot learning

Graph augmentation for node-level few-shot learning

In graph few-shot learning, few-shot node classification (FSNC) at the node-level is a popular downstream task. Previous FSNC methods primarily rely on meta-learning or metric learning techniques, aiming to mine prior knowledge from the base classes. However, these methods still have some limitations that need to be addressed, namely: (1) conducting multiple tasks for parameter initialization leads to expensive time costs. (2) ignoring the rich information present in novel classes leads to model over-fitting. To address these issues, this paper proposes a novel graph augmentation method for FSNC on graph data, which includes both parameter initialization and parameter fine-tuning. Specifically, the parameter initialization conducts only one multi-classification task on the base classes, improving generalization ability and reducing time costs. The parameter fine-tuning is designed to include two data augmentation modules (i.e., support augmentation and shot augmentation) on the novel classes to mine the rich information, thus alleviating model over-fitting. As a result, this paper introduces the first graph augmentation method for FSNC. Experimental results showed that our method achieves supreme performance, compared with state-of-the-art FSNC methods.

Semi-supervised few-shot class-incremental learning based on dynamic topology evolution

Incremental learning of new classes is crucial to developing real-world artificial intelligence systems. However, in some cases, we can only use a limited number of images to train the incremental model due to the difficulty of obtaining them. Few-shot class-incremental learning (FSCIL) is a new paradigm of incremental learning, which continually learns new concepts from a few examples. This setup can quickly lead to catastrophic forgetting and overfitting problems, severely affecting model performance. To address these problems, we propose a dynamic topological evolution (DTE) model that unifies knowledge of old and new categories. In this work, we focus on a novel semi-supervised few-shot class-incremental learning (SSFSCIL) to incrementally learn new categories with a limited number of labeled samples and a large number of unlabeled samples. In detail, we use the Voronoi polygon criterion to divide the feature space and construct a Hebbian graph to represent topological relationships between classes. Then, the corresponding k-nearest neighbor subgraph is constructed for the new incremental task, and label propagation of graph nodes is carried out on the unlabeled set to generate the corresponding pseudo labels. Finally, the new category knowledge is evolved into the actual category knowledge by integrating the Hebbian graph and the k-nearest neighbor subgraph obtained in the new incremental task. Extensive experiments on the mini-ImageNet and CIFAR100 datasets show that the accuracy of the DTE outperforms the baseline and improves by 1.41% and 1.13% on the last incremental task. Furthermore, our methodology achieved a 35.6% time improvement compared to full-supervised methods on the CIFAR100 dataset. In addition to the experiments on the two datasets mentioned above, the validation of our proposed DTE model extends to the engineering application domain of the RESICS-45 aircraft image scene classification dataset.

Few-shot learning and modeling of 3D reservoir properties for predicting oil reservoir production

Few-shot learning and modeling of 3D reservoir properties for predicting oil reservoir production

Modal-aware prompt tuning with deep adaptive feature enhancement

Prompt learning has recently emerged as a promising method for fine-tuning vision-language models. By introducing prompts in the text encoder or image encoder, the pre-trained model can quickly adapt to downstream tasks without updating the pre-trained weights. However, prior multi-modal prompt tuning works do not consider the difference in feature distributions between text and images, and adopt the same prompts for both encoders, thus achieving sub-optimal performance in the downstream few-shot learning. In this paper, we propose Modal-Aware Prompt (MAP) to alleviate this issue. Specifically, considering the stability of text features, we design text-specific prompts, which can acquire text class-related information from a general template (i.e., “a photo of a <category>”) by unidirectional attention-based interaction. Additionally, considering the diversity of image features, we design visual-specific prompts to acquire image class-related information and adjust the image features by bidirectional attention-based interaction. To learn hierarchical prompt representations and reinforce the prompt features, we further propose a Deep Adaptive Feature Enhancement (DAFE) module to adaptively utilize the prompt output of the former layer, which can combine instance-level and task-level information simultaneously. Combining the above two designs, our method MAP-DAFE obtains state-of-the-art results on 11 image recognition datasets and has the fastest convergence rate. This proves our MAP-DAFE is effective and efficient.

Multimodal Sentiment Analysis for Movie Scenes Based on a Few-Shot Learning Approach

Multimodal sentiment analysis involves discerning the emotions of speakers through diverse features such as sound, text, and images. However, current research in this field heavily relies on extensive supervised datasets, demanding substantial labor and computational resources. This study introduces a meta-learning-based approach for multimodal sentiment analysis, aiming to delve into emotional information within movie scenes. Leveraging meta-learning techniques, this approach seeks to accurately capture emotions in movie scenes using a limited number of annotated samples, achieving model generalization under constrained labeled data. Specifically, we introduce an optimization-based meta-learning approach for the multimodal sentiment analysis tasks in text and vision, enhancing the model’s ability to generalize to new tasks with limited annotations. Additionally, an innovative strategy based on Meta-Prompting is proposed to handle multimodal data. The merits of the proposed method are validated across different datasets.

Harnessing large language models over transformer models for detecting Bengali depressive social media text: A comprehensive study

In an era where the silent struggle of underdiagnosed depression pervades globally, our research delves into the crucial link between mental health and social media. This work focuses on early detection of depression, particularly in extroverted social media users, using LLMs such as GPT 3.5, GPT 4 and our proposed GPT 3.5 fine-tuned model DepGPT, as well as advanced Deep learning models(LSTM, Bi-LSTM, GRU, BiGRU) and Transformer models(BERT, BanglaBERT, SahajBERT, BanglaBERT-Base). The study categorized Reddit and X datasets into “Depressive” and “Non-Depressive” segments, translated into Bengali by native speakers with expertise in mental health, resulting in the creation of the Bengali Social Media Depressive Dataset (BSMDD). Our work provides full architecture details for each model and a methodical way to assess their performance in Bengali depressive text categorization using zero-shot and few-shot learning techniques. Our work demonstrates the superiority of SahajBERT and Bi-LSTM with FastText embeddings in their respective domains also tackles explainability issues with transformer models and emphasizes the effectiveness of LLMs, especially DepGPT (GPT 3.5 fine-tuned), demonstrating flexibility and competence in a range of learning contexts. According to the experiment results, the proposed model, DepGPT, outperformed not only Alpaca Lora 7B in zero-shot and few-shot scenarios but also every other model, achieving a near-perfect accuracy of 0.9796 and an F1-score of 0.9804, high recall, and exceptional precision. Although competitive, GPT-3.5 Turbo and Alpaca Lora 7B show relatively poorer effectiveness in zero-shot and few-shot situations. The work emphasizes the effectiveness and flexibility of LLMs in a variety of linguistic circumstances, providing insightful information about the complex field of depression detection models.

Few-Shot Federated Learning: A Federated Learning Model for Small-Sample Scenarios

Traditional federated learning relies heavily on mature datasets, which typically consist of large volumes of uniformly distributed data. While acquiring extensive datasets is relatively straightforward in academic research, it becomes prohibitively expensive in practical applications, especially in emerging or specialized medical fields characterized by data scarcity. This poses a significant challenge. To address this issue, our study introduces a federated learning model that integrates few-shot learning techniques and is complemented by personalized knowledge distillation to further enhance the model’s classification accuracy. This innovative approach significantly reduces the dependence on large-scale datasets, enabling efficient model training under limited data conditions. Our experimental evaluations conducted on small-scale datasets, including Omniglot, FC100, and mini-ImageNet, indicate that our model surpasses existing state-of-the-art federated learning models in terms of accuracy, achieving a substantial improvement. Specifically, on the FC100 dataset, the classification accuracy of the conventional federated learning algorithm FedAvg was merely 19.6%, whereas the method proposed in this study achieved a classification accuracy of 41%, representing an improvement of more than double. This advancement not only highlights our model’s superiority in alleviating the challenges of limited data availability, but also expands the applicability of federated learning to a broader range of applications.