RESUMO
Humans and animals excel at generalizing from limited data, a capability yet to be fully replicated in artificial intelligence. This perspective investigates generalization in biological and artificial deep neural networks (DNNs), in both in-distribution and out-of-distribution contexts. We introduce two hypotheses: First, the geometric properties of the neural manifolds associated with discrete cognitive entities, such as objects, words, and concepts, are powerful order parameters. They link the neural substrate to the generalization capabilities and provide a unified methodology bridging gaps between neuroscience, machine learning, and cognitive science. We overview recent progress in studying the geometry of neural manifolds, particularly in visual object recognition, and discuss theories connecting manifold dimension and radius to generalization capacity. Second, we suggest that the theory of learning in wide DNNs, especially in the thermodynamic limit, provides mechanistic insights into the learning processes generating desired neural representational geometries and generalization. This includes the role of weight norm regularization, network architecture, and hyper-parameters. We will explore recent advances in this theory and ongoing challenges. We also discuss the dynamics of learning and its relevance to the issue of representational drift in the brain.
Assuntos
Encéfalo , Redes Neurais de Computação , Encéfalo/fisiologia , Humanos , Animais , Inteligência Artificial , Modelos Neurológicos , Generalização Psicológica/fisiologia , Cognição/fisiologiaRESUMO
The molecular property prediction (MPP) plays a crucial role in the drug discovery process, providing valuable insights for molecule evaluation and screening. Although deep learning has achieved numerous advances in this area, its success often depends on the availability of substantial labeled data. The few-shot MPP is a more challenging scenario, which aims to identify unseen property with only few available molecules. In this paper, we propose an attribute-guided prototype network (APN) to address the challenge. APN first introduces an molecular attribute extractor, which can not only extract three different types of fingerprint attributes (single fingerprint attributes, dual fingerprint attributes, triplet fingerprint attributes) by considering seven circular-based, five path-based, and two substructure-based fingerprints, but also automatically extract deep attributes from self-supervised learning methods. Furthermore, APN designs the Attribute-Guided Dual-channel Attention module to learn the relationship between the molecular graphs and attributes and refine the local and global representation of the molecules. Compared with existing works, APN leverages high-level human-defined attributes and helps the model to explicitly generalize knowledge in molecular graphs. Experiments on benchmark datasets show that APN can achieve state-of-the-art performance in most cases and demonstrate that the attributes are effective for improving few-shot MPP performance. In addition, the strong generalization ability of APN is verified by conducting experiments on data from different domains.
Assuntos
Aprendizado Profundo , Descoberta de Drogas , Descoberta de Drogas/métodos , Humanos , Algoritmos , Redes Neurais de ComputaçãoRESUMO
Large language models (LLMs) are sophisticated AI-driven models trained on vast sources of natural language data. They are adept at generating responses that closely mimic human conversational patterns. One of the most notable examples is OpenAI's ChatGPT, which has been extensively used across diverse sectors. Despite their flexibility, a significant challenge arises as most users must transmit their data to the servers of companies operating these models. Utilizing ChatGPT or similar models online may inadvertently expose sensitive information to the risk of data breaches. Therefore, implementing LLMs that are open source and smaller in scale within a secure local network becomes a crucial step for organizations where ensuring data privacy and protection has the highest priority, such as regulatory agencies. As a feasibility evaluation, we implemented a series of open-source LLMs within a regulatory agency's local network and assessed their performance on specific tasks involving extracting relevant clinical pharmacology information from regulatory drug labels. Our research shows that some models work well in the context of few- or zero-shot learning, achieving performance comparable, or even better than, neural network models that needed thousands of training samples. One of the models was selected to address a real-world issue of finding intrinsic factors that affect drugs' clinical exposure without any training or fine-tuning. In a dataset of over 700 000 sentences, the model showed a 78.5% accuracy rate. Our work pointed to the possibility of implementing open-source LLMs within a secure local network and using these models to perform various natural language processing tasks when large numbers of training examples are unavailable.
Assuntos
Processamento de Linguagem Natural , Humanos , Redes Neurais de Computação , Aprendizado de MáquinaRESUMO
MOTIVATION: Histones are the chief protein components of chromatin, and the chemical modifications on histones crucially influence the transcriptional state of related genes. Histone modifying enzyme (HME), responsible for adding or removing the chemical labels, has emerged as a very important class of drug target, with a few HME inhibitors launched as anti-cancerous drugs and tens of molecules under clinical trials. To accelerate the drug discovery process of HME inhibitors, machine learning-based predictive models have been developed to enrich the active molecules from vast chemical space. However, the number of compounds with known activity distributed largely unbalanced among different HMEs, particularly with many targets of less than a hundred active samples. In this case, it is difficult to build effective virtual screening models directly based on machine learning. RESULTS: To this end, we propose a new Meta-learning-based Histone Modifying Enzymes Inhibitor prediction method (MetaHMEI). Our proposed MetaHMEI first uses a self-supervised pre-training approach to obtain high-quality molecular substructure embeddings from a large unlabeled chemical dataset. Then, MetaHMEI exploits a Transformer-based encoder and meta-learning framework to build a prediction model. MetaHMEI allows the effective transfer of the prior knowledge learned from HMEs with sufficient samples to HMEs with a small number of samples, so the proposed model can produce accurate predictions for HMEs with limited data. Extensive experimental results on our collected and curated HMEs datasets show that MetaHMEI is better than other methods in the case of few-shot learning. Furthermore, we applied MetaHMEI in the virtual screening process of histone JMJD3 inhibitors and successfully obtained three small molecule inhibitors, further supporting the validity of our model.
Assuntos
Cromatina , Histonas , Histonas/metabolismo , Descoberta de Drogas/métodos , Inibidores Enzimáticos/farmacologiaRESUMO
Understanding the neural basis of the remarkable human cognitive capacity to learn novel concepts from just one or a few sensory experiences constitutes a fundamental problem. We propose a simple, biologically plausible, mathematically tractable, and computationally powerful neural mechanism for few-shot learning of naturalistic concepts. We posit that the concepts that can be learned from few examples are defined by tightly circumscribed manifolds in the neural firing-rate space of higher-order sensory areas. We further posit that a single plastic downstream readout neuron learns to discriminate new concepts based on few examples using a simple plasticity rule. We demonstrate the computational power of our proposal by showing that it can achieve high few-shot learning accuracy on natural visual concepts using both macaque inferotemporal cortex representations and deep neural network (DNN) models of these representations and can even learn novel visual concepts specified only through linguistic descriptors. Moreover, we develop a mathematical theory of few-shot learning that links neurophysiology to predictions about behavioral outcomes by delineating several fundamental and measurable geometric properties of neural representations that can accurately predict the few-shot learning performance of naturalistic concepts across all our numerical simulations. This theory reveals, for instance, that high-dimensional manifolds enhance the ability to learn new concepts from few examples. Intriguingly, we observe striking mismatches between the geometry of manifolds in the primate visual pathway and in trained DNNs. We discuss testable predictions of our theory for psychophysics and neurophysiological experiments.
Assuntos
Formação de Conceito , Redes Neurais de Computação , Animais , Humanos , Aprendizagem/fisiologia , Macaca , Plásticos , Primatas , Vias Visuais/fisiologiaRESUMO
Plant diseases worsen the threat of food shortage with the growing global population, and disease recognition is the basis for the effective prevention and control of plant diseases. Deep learning has made significant breakthroughs in the field of plant disease recognition. Compared with traditional deep learning, meta-learning can still maintain more than 90% accuracy in disease recognition with small samples. However, there is no comprehensive review on the application of meta-learning in plant disease recognition. Here, we mainly summarize the functions, advantages, and limitations of meta-learning research methods and their applications for plant disease recognition with a few data scenarios. Finally, we outline several research avenues for utilizing current and future meta-learning in plant science. This review may help plant science researchers obtain faster, more accurate, and more credible solutions through deep learning with fewer labeled samples.
Assuntos
Doenças das Plantas , Aprendizado ProfundoRESUMO
Drug-drug interactions (DDIs) are one of the major concerns in pharmaceutical research, and a number of computational methods have been developed to predict whether two drugs interact or not. Recently, more attention has been paid to events caused by the DDIs, which is more useful for investigating the mechanism hidden behind the combined drug usage or adverse reactions. However, some rare events may only have few examples, hindering them from being precisely predicted. To address the above issues, we present a few-shot computational method named META-DDIE, which consists of a representation module and a comparing module, to predict DDI events. We collect drug chemical structures and DDIs from DrugBank, and categorize DDI events into hundreds of types using a standard pipeline. META-DDIE uses the structures of drugs as input and learns the interpretable representations of DDIs through the representation module. Then, the model uses the comparing module to predict whether two representations are similar, and finally predicts DDI events with few labeled examples. In the computational experiments, META-DDIE outperforms several baseline methods and especially enhances the predictive capability for rare events. Moreover, META-DDIE helps to identify the key factors that may cause DDI events and reveal the relationship among different events.
Assuntos
Interações Medicamentosas , Preparações Farmacêuticas , Bases de Dados Factuais , Efeitos Colaterais e Reações Adversas Relacionados a Medicamentos , Humanos , Modelos TeóricosRESUMO
Recently, machine learning methods have been developed to identify various peptide bio-activities. However, due to the lack of experimentally validated peptides, machine learning methods cannot provide a sufficiently trained model, easily resulting in poor generalizability. Furthermore, there is no generic computational framework to predict the bioactivities of different peptides. Thus, a natural question is whether we can use limited samples to build an effective predictive model for different kinds of peptides. To address this question, we propose Mutual Information Maximization Meta-Learning (MIMML), a novel meta-learning-based predictive model for bioactive peptide discovery. Using few samples from various functional peptides, MIMML can sufficiently learn the discriminative information amongst various functions and characterize functional differences. Experimental results show excellent performance of MIMML though using far fewer training samples as compared to the state-of-the-art methods. We also decipher the latent relationships among different kinds of functions to understand what meta-model learned to improve a specific task. In summary, this study is a pioneering work in the field of functional peptide mining and provides the first-of-its-kind solution for few-sample learning problems in biological sequence analysis, accelerating the new functional peptide discovery. The source codes and datasets are available on https://github.com/TearsWaiting/MIMML.
Assuntos
Aprendizado de Máquina , Peptídeos , Peptídeos/química , SoftwareRESUMO
AIMS: Recent advances in artificial intelligence, particularly with large language models like GPT-4Vision (GPT-4V)-a derivative feature of ChatGPT-have expanded the potential for medical image interpretation. This study evaluates the accuracy of GPT-4V in image classification tasks of histopathological images and compares its performance with a traditional convolutional neural network (CNN). METHODS: We utilised 1520 images, including haematoxylin and eosin staining and tau immunohistochemistry, from patients with various neurodegenerative diseases, such as Alzheimer's disease (AD), progressive supranuclear palsy (PSP) and corticobasal degeneration (CBD). We assessed GPT-4V's performance using multi-step prompts to determine how textual context influences image interpretation. We also employed few-shot learning to enhance improvements in GPT-4V's diagnostic performance in classifying three specific tau lesions-astrocytic plaques, neuritic plaques and tufted astrocytes-and compared the outcomes with the CNN model YOLOv8. RESULTS: GPT-4V accurately recognised staining techniques and tissue origin but struggled with specific lesion identification. The interpretation of images was notably influenced by the provided textual context, which sometimes led to diagnostic inaccuracies. For instance, when presented with images of the motor cortex, the diagnosis shifted inappropriately from AD to CBD or PSP. However, few-shot learning markedly improved GPT-4V's diagnostic capabilities, enhancing accuracy from 40% in zero-shot learning to 90% with 20-shot learning, matching the performance of YOLOv8, which required 100-shot learning to achieve the same accuracy. CONCLUSIONS: Although GPT-4V faces challenges in independently interpreting histopathological images, few-shot learning significantly improves its performance. This approach is especially promising for neuropathology, where acquiring extensive labelled datasets is often challenging.
Assuntos
Redes Neurais de Computação , Doenças Neurodegenerativas , Humanos , Doenças Neurodegenerativas/patologia , Interpretação de Imagem Assistida por Computador/métodos , Doença de Alzheimer/patologiaRESUMO
Magnetic resonance imaging (MRI) is a ubiquitous medical imaging technology with applications in disease diagnostics, intervention, and treatment planning. Accurate MRI segmentation is critical for diagnosing abnormalities, monitoring diseases, and deciding on a course of treatment. With the advent of advanced deep learning frameworks, fully automated and accurate MRI segmentation is advancing. Traditional supervised deep learning techniques have advanced tremendously, reaching clinical-level accuracy in the field of segmentation. However, these algorithms still require a large amount of annotated data, which is oftentimes unavailable or impractical. One way to circumvent this issue is to utilize algorithms that exploit a limited amount of labeled data. This paper aims to review such state-of-the-art algorithms that use a limited number of annotated samples. We explain the fundamental principles of self-supervised learning, generative models, few-shot learning, and semi-supervised learning and summarize their applications in cardiac, abdomen, and brain MRI segmentation. Throughout this review, we highlight algorithms that can be employed based on the quantity of annotated data available. We also present a comprehensive list of notable publicly available MRI segmentation datasets. To conclude, we discuss possible future directions of the field-including emerging algorithms, such as contrastive language-image pretraining, and potential combinations across the methods discussed-that can further increase the efficacy of image segmentation with limited labels.
Assuntos
Aprendizado Profundo , Imageamento por Ressonância Magnética , Humanos , Imageamento por Ressonância Magnética/métodos , Processamento de Imagem Assistida por Computador/métodos , Algoritmos , Aprendizado de Máquina Supervisionado , Encéfalo/diagnóstico por imagemRESUMO
BACKGROUND: In-home rehabilitation systems are a promising, potential alternative to conventional therapy for stroke survivors. Unfortunately, physiological differences between participants and sensor displacement in wearable sensors pose a significant challenge to classifier performance, particularly for people with stroke who may encounter difficulties repeatedly performing trials. This makes it challenging to create reliable in-home rehabilitation systems that can accurately classify gestures. METHODS: Twenty individuals who suffered a stroke performed seven different gestures (mass flexion, mass extension, wrist volar flexion, wrist dorsiflexion, forearm pronation, forearm supination, and rest) related to activities of daily living. They performed these gestures while wearing EMG sensors on the forearm, as well as FMG sensors and an IMU on the wrist. We developed a model based on prototypical networks for one-shot transfer learning, K-Best feature selection, and increased window size to improve model accuracy. Our model was evaluated against conventional transfer learning with neural networks, as well as subject-dependent and subject-independent classifiers: neural networks, LGBM, LDA, and SVM. RESULTS: Our proposed model achieved 82.2% hand-gesture classification accuracy, which was better (P<0.05) than one-shot transfer learning with neural networks (63.17%), neural networks (59.72%), LGBM (65.09%), LDA (63.35%), and SVM (54.5%). In addition, our model performed similarly to subject-dependent classifiers, slightly lower than SVM (83.84%) but higher than neural networks (81.62%), LGBM (80.79%), and LDA (74.89%). Using K-Best features improved the accuracy in 3 of the 6 classifiers used for evaluation, while not affecting the accuracy in the other classifiers. Increasing the window size improved the accuracy of all the classifiers by an average of 4.28%. CONCLUSION: Our proposed model showed significant improvements in hand-gesture recognition accuracy in individuals who have had a stroke as compared with conventional transfer learning, neural networks and traditional machine learning approaches. In addition, K-Best feature selection and increased window size can further improve the accuracy. This approach could help to alleviate the impact of physiological differences and create a subject-independent model for stroke survivors that improves the classification accuracy of wearable sensors. TRIAL REGISTRATION NUMBER: The study was registered in Chinese Clinical Trial Registry with registration number CHiCTR1800017568 in 2018/08/04.
Assuntos
Gestos , Mãos , Redes Neurais de Computação , Reabilitação do Acidente Vascular Cerebral , Humanos , Reabilitação do Acidente Vascular Cerebral/métodos , Reabilitação do Acidente Vascular Cerebral/instrumentação , Mãos/fisiopatologia , Masculino , Feminino , Pessoa de Meia-Idade , Acidente Vascular Cerebral/complicações , Acidente Vascular Cerebral/fisiopatologia , Idoso , Aprendizado de Máquina , Transferência de Experiência/fisiologia , Adulto , Eletromiografia , Dispositivos Eletrônicos VestíveisRESUMO
Few-shot object detection is a challenging task aimed at recognizing novel classes and localizing with limited labeled data. Although substantial achievements have been obtained, existing methods mostly struggle with forgetting and lack stability across various few-shot training samples. In this paper, we reveal two gaps affecting meta-knowledge transfer, leading to unstable performance and forgetting in meta-learning-based frameworks. To this end, we propose sample normalization, a simple yet effective method that enhances performance stability and decreases forgetting. Additionally, we apply Z-score normalization to mitigate the hubness problem in high-dimensional feature space. Experimental results on the PASCAL VOC data set demonstrate that our approach outperforms existing methods in both accuracy and stability, achieving up to +4.4 mAP@0.5 and +5.3 mAR in a single run, with +4.8 mAP@0.5 and +5.1 mAR over 10 random experiments on average. Furthermore, our method alleviates the drop in performance of base classes. The code will be released to facilitate future research.
RESUMO
Large vision-language models, such as Contrastive Vision-Language Pre-training (CLIP), pre-trained on large-scale image-text datasets, have demonstrated robust zero-shot transfer capabilities across various downstream tasks. To further enhance the few-shot recognition performance of CLIP, Tip-Adapter augments the CLIP model with an adapter that incorporates a key-value cache model constructed from the few-shot training set. This approach enables training-free adaptation and has shown significant improvements in few-shot recognition, especially with additional fine-tuning. However, the size of the adapter increases in proportion to the number of training samples, making it difficult to deploy in practical applications. In this paper, we propose a novel CLIP adaptation method, named Proto-Adapter, which employs a single-layer adapter of constant size regardless of the amount of training data and even outperforms Tip-Adapter. Proto-Adapter constructs the adapter's weights based on prototype representations for each class. By aggregating the features of the training samples, it successfully reduces the size of the adapter without compromising performance. Moreover, the performance of the model can be further enhanced by fine-tuning the adapter's weights using a distance margin penalty, which imposes additional inter-class discrepancy to the output logits. We posit that this training scheme allows us to obtain a model with a discriminative decision boundary even when trained with a limited amount of data. We demonstrate the effectiveness of the proposed method through extensive experiments of few-shot classification on diverse datasets.
RESUMO
Significant strides have been made in the field of WiFi-based human activity recognition, yet recent wireless sensing methodologies still grapple with the reliance on copious amounts of data. When assessed in unfamiliar domains, the majority of models experience a decline in accuracy. To address this challenge, this study introduces Wi-CHAR, a novel few-shot learning-based cross-domain activity recognition system. Wi-CHAR is meticulously designed to tackle both the intricacies of specific sensing environments and pertinent data-related issues. Initially, Wi-CHAR employs a dynamic selection methodology for sensing devices, tailored to mitigate the diminished sensing capabilities observed in specific regions within a multi-WiFi sensor device ecosystem, thereby augmenting the fidelity of sensing data. Subsequent refinement involves the utilization of the MF-DBSCAN clustering algorithm iteratively, enabling the rectification of anomalies and enhancing the quality of subsequent behavior recognition processes. Furthermore, the Re-PN module is consistently engaged, dynamically adjusting feature prototype weights to facilitate cross-domain activity sensing in scenarios with limited sample data, effectively distinguishing between accurate and noisy data samples, thus streamlining the identification of new users and environments. The experimental results show that the average accuracy is more than 93% (five-shot) in various scenarios. Even in cases where the target domain has fewer data samples, better cross-domain results can be achieved. Notably, evaluation on publicly available datasets, WiAR and Widar 3.0, corroborates Wi-CHAR's robust performance, boasting accuracy rates of 89.7% and 92.5%, respectively. In summary, Wi-CHAR delivers recognition outcomes on par with state-of-the-art methodologies, meticulously tailored to accommodate specific sensing environments and data constraints.
RESUMO
Automatic Modulation Recognition (AMR) is a key technology in the field of cognitive communication, playing a core role in many applications, especially in wireless security issues. Currently, deep learning (DL)-based AMR technology has achieved many research results, greatly promoting the development of AMR technology. However, the few-shot dilemma faced by DL-based AMR methods greatly limits their application in practical scenarios. Therefore, this paper endeavored to address the challenge of AMR with limited data and proposed a novel meta-learning method, the Multi-Level Comparison Relation Network with Class Reconstruction (MCRN-CR). Firstly, the method designs a structure of a multi-level comparison relation network, which involves embedding functions to output their feature maps hierarchically, comprehensively calculating the relation scores between query samples and support samples to determine the modulation category. Secondly, the embedding function integrates a reconstruction module, leveraging an autoencoder for support sample reconstruction, wherein the encoder serves dual purposes as the embedding mechanism. The training regimen incorporates a meta-learning paradigm, harmoniously combining classification and reconstruction losses to refine the model's performance. The experimental results on the RadioML2018 dataset show that our designed method can greatly alleviate the small sample problem in AMR and is superior to existing methods.
RESUMO
Dynamic visual vending machines are rapidly growing in popularity, offering convenience and speed to customers. However, there is a prevalent issue with consumers damaging goods and then returning them to the machine, severely affecting business interests. This paper addresses the issue from the standpoint of defect detection. Although existing industrial defect detection algorithms, such as PatchCore, perform well, they face challenges, including handling goods in various orientations, detection speeds that do not meet real-time monitoring requirements, and complex backgrounds that hinder detection accuracy. These challenges hinder their application in dynamic vending environments. It is crucial to note that efficient visual features play a vital role in memory banks, yet current memory repositories for industrial inspection algorithms do not adequately address the problem of location-specific feature redundancy. To tackle these issues, this paper introduces a novel defect detection algorithm for goods using adaptive subsampling and partitioned memory banks. Firstly, Grad-CAM is utilized to extract deep features, which, in combination with shallow features, mitigate the impact of complex backgrounds on detection accuracy. Next, graph convolutional networks extract rotationally invariant features. The adaptive subsampling partitioned memory bank is then employed to store features of non-defective goods, which reduces memory consumption and enhances training speed. Experimental results on the MVTec AD dataset demonstrate that the proposed algorithm achieves a marked improvement in detection speed while maintaining accuracy that is comparable to state-of-the-art models.
RESUMO
The extraction of effective classification features from high-dimensional hyperspectral images, impeded by the scarcity of labeled samples and uneven sample distribution, represents a formidable challenge within hyperspectral image classification. Traditional few-shot learning methods confront the dual dilemma of limited annotated samples and the necessity for deeper, more effective features from complex hyperspectral data, often resulting in suboptimal outcomes. The prohibitive cost of sample annotation further exacerbates the challenge, making it difficult to rely on a scant number of annotated samples for effective feature extraction. Prevailing high-accuracy algorithms require abundant annotated samples and falter in deriving deep, discriminative features from limited data, compromising classification performance for complex substances. This paper advocates for an integration of advanced spectral-spatial feature extraction with meta-transfer learning to address the classification of hyperspectral signals amidst insufficient labeled samples. Initially trained on a source domain dataset with ample labels, the model undergoes transference to a target domain with minimal samples, utilizing dense connection blocks and tree-dimensional convolutional residual connections to enhance feature extraction and maximize spatial and spectral information retrieval. This approach, validated on three diverse hyperspectral datasets-IP, UP, and Salinas-significantly surpasses existing classification algorithms and small-sample techniques in accuracy, demonstrating its applicability to high-dimensional signal classification under label constraints.
RESUMO
Supervised learning methods excel in traditional relation extraction tasks. However, the quality and scale of the training data heavily influence their performance. Few-shot relation extraction is gradually becoming a research hotspot whose objective is to learn and extract semantic relationships between entities with only a limited number of annotated samples. In recent years, numerous studies have employed prototypical networks for few-shot relation extraction. However, these methods often suffer from overfitting of the relation classes, making it challenging to generalize effectively to new relationships. Therefore, this paper seeks to utilize a diffusion model for data augmentation to address the overfitting issue of prototypical networks. We propose a diffusion model-enhanced prototypical network framework. Specifically, we design and train a controllable conditional relation generation diffusion model on the relation extraction dataset, which can generate the corresponding instance representation according to the relation description. Building upon the trained diffusion model, we further present a pseudo-sample-enhanced prototypical network, which is able to provide more accurate representations for prototype classes, thereby alleviating overfitting and better generalizing to unseen relation classes. Additionally, we introduce a pseudo-sample-aware attention mechanism to enhance the model's adaptability to pseudo-sample data through a cross-entropy loss, further improving the model's performance. A series of experiments are conducted to prove our method's effectiveness. The results indicate that our proposed approach significantly outperforms existing methods, particularly in low-resource one-shot environments. Further ablation analyses underscore the necessity of each module in the model. As far as we know, this is the first research to employ a diffusion model for enhancing the prototypical network through data augmentation in few-shot relation extraction.
RESUMO
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.
RESUMO
Few-shot learning, especially few-shot image classification (FSIC), endeavors to recognize new categories using only a handful of labeled images by transferring knowledge from a model trained on base categories. Despite numerous efforts to address the challenge of deficient transferability caused by the distribution shift between the base and new classes, the fundamental principles remain a subject of debate. In this paper, we elucidate why a decline in performance occurs and what information is transferred during the testing phase, examining it from a frequency spectrum perspective. Specifically, we adopt causality on the frequency space for FSIC. With our causal assumption, non-causal frequencies (e.g., background knowledge) act as confounders between causal frequencies (e.g., object information) and predictions. Our experimental results reveal that different frequency components represent distinct semantics, and non-causal frequencies adversely affect transferability, resulting in suboptimal performance. Subsequently, we suggest a straightforward but potent approach, namely the Frequency Spectrum Mask (FRSM), to weight the frequency and mitigate the impact of non-causal frequencies. Extensive experiments demonstrate that the proposed FRSM method significantly enhanced the transferability of the FSIC model across nine testing datasets.