Sample Labeling Research Articles

A self-calibration model building method for rapid total nitrogen detection based on semi-supervised learning

In practical TN detection, inaccurate labels, sample noises, and the insufficiency of labeled samples are the three most predominant problems which degrade the detection performance. In this paper, we proposed a novel anti-noise semi-supervised learning method called multi-regularized robustness semi-supervised learning (MR2S2L), which can establish accurate detection models based on inaccurate and insufficient labeled samples. The strategy of structured sparse learning and adaptive graph learning are integrated into MR2S2L firstly to reduce the negative influence brought by sample noises. Then, a multi-regularized calibration method is proposed to deal with the noising label problem, which modifies the inaccurate labels based on the spectrum relationships and temporal correlations during the sample-collecting process. Experimental results based on two practical datasets show that compared with traditional supervised learning methods, MR2S2L can reduce the size of labeled samples to the original 10 % and improve the detection accuracy to twice the original.

Variational deep embedding-based active learning for the diagnosis of pneumonia.

Machine learning works similar to the way humans train their brains. In general, previous experiences prepared the brain by firing specific nerve cells in the brain and increasing the weight of the links between them. Machine learning also completes the classification task by constantly changing the weights in the model through training on the training set. It can conduct a much more significant amount of training and achieve higher recognition accuracy in specific fields than the human brain. In this paper, we proposed an active learning framework called variational deep embedding-based active learning (VaDEAL) as a human-centric computing method to improve the accuracy of diagnosing pneumonia. Because active learning (AL) realizes label-efficient learning by labeling the most valuable queries, we propose a new AL strategy that incorporates clustering to improve the sampling quality. Our framework consists of a VaDE module, a task learner, and a sampling calculator. First, the VaDE performs unsupervised reduction and clustering of dimension over the entire data set. The end-to-end task learner obtains the embedding representations of the VaDE-processed sample while training the target classifier of the model. The sampling calculator will calculate the representativeness of the samples by VaDE, the uncertainty of the samples through task learning, and ensure the overall diversity of the samples by calculating the similarity constraints between the current and previous samples. With our novel design, the combination of uncertainty, representativeness, and diversity scores allows us to select the most informative samples for labeling, thus improving overall performance. With extensive experiments and evaluations performed on a large dataset, we demonstrate that our proposed method is superior to the state-of-the-art methods and has the highest accuracy in the diagnosis of pneumonia.

Deep Learning-Based Cyber–Physical Feature Fusion for Anomaly Detection in Industrial Control Systems

In this paper, we propose an unsupervised anomaly detection method based on the Autoencoder with Long Short-Term Memory (LSTM-Autoencoder) network and Generative Adversarial Network (GAN) to detect anomalies in industrial control system (ICS) using cyber–physical fusion features. This method improves the recall of anomaly detection and overcomes the challenges of unbalanced datasets and insufficient labeled samples in ICS. As a first step, additional network features are extracted and fused with physical features to create a cyber–physical dataset. Following this, the model is trained using normal data to ensure that it can properly reconstruct the normal data. In the testing phase, samples with unknown labels are used as inputs to the model. The model will output an anomaly score for each sample, and whether a sample is anomalous depends on whether the anomaly score exceeds the threshold. Whether using supervised or unsupervised algorithms, experimentation has shown that (1) cyber–physical fusion features can significantly improve the performance of anomaly detection algorithms; (2) the proposed method outperforms several other unsupervised anomaly detection methods in terms of accuracy, recall, and F1 score; (3) the proposed method can detect the majority of anomalous events with a low false negative rate.

A New SCAE-MT Classification Model for Hyperspectral Remote Sensing Images.

Hyperspectral remote sensing images (HRSI) have the characteristics of foreign objects with the same spectrum. As it is difficult to label samples manually, the hyperspectral remote sensing images are understood to be typical "small sample" datasets. Deep neural networks can effectively extract the deep features from the HRSI, but the classification accuracy mainly depends on the training label samples. Therefore, the stacked convolutional autoencoder network and transfer learning strategy are employed in order to design a new stacked convolutional autoencoder network model transfer (SCAE-MT) for the purposes of classifying the HRSI in this paper. In the proposed classification method, the stacked convolutional au-to-encoding network is employed in order to effectively extract the deep features from the HRSI. Then, the transfer learning strategy is applied to design a stacked convolutional autoencoder network model transfer under the small and limited training samples. The SCAE-MT model is used to propose a new HRSI classification method in order to solve the small samples of the HRSI. In this study, in order to prove the effectiveness of the proposed classification method, two HRSI datasets were chosen. In order to verify the effectiveness of the methods, the overall classification accuracy (OA) of the convolutional self-coding network classification method (CAE), the stack convolutional self-coding network classification method (SCAE), and the SCAE-MT method under 5%, 10%, and 15% training sets are calculated. When compared with the CAE and SCAE models in 5%, 10%, and 15% training datasets, the overall accuracy (OA) of the SCAE-MT method was improved by 2.71%, 3.33%, and 3.07% (on average), respectively. The SCAE-MT method is, thus, clearly superior to the other methods and also shows a good classification performance.

A Semi-Supervised Inspection Approach of Textured Surface Defects under Limited Labeled Samples

Defect inspection is a key step in guaranteeing the surface quality of industrial products. Based on deep learning (DL) techniques, related methods are highly effective in defect classification tasks via a supervision process. However, collecting and labeling many defect samples are usually harsh and time-consuming processes, limiting the application of these supervised classifiers on various textured surfaces. This study proposes a semi-supervised framework, based on a generative adversarial network (GAN) and a convolutional neural network (CNN), to classify defects of a textured surface, while a novel label assignment scheme is proposed to integrate unlabeled samples into semi-supervised learning to enhance the overall performance of the system. In this framework, a customized GAN uses limited labeled samples to generate unlabeled ones, while the proposed label assignment scheme makes the generated data follow different label distributions in such a way that they can participate in training with labeled data. Finally, a CNN is proposed for semi-supervised training and the category identification of each defect sample. Experimental results show the effectiveness and robustness of the proposed framework even if original samples are limited. We verify our approach on four different surface defect datasets, achieving consistently competitive performances.

VQAMix: Conditional Triplet Mixup for Medical Visual Question Answering

Medical visual question answering (VQA) aims to correctly answer a clinical question related to a given medical image. Nevertheless, owing to the expensive manual annotations of medical data, the lack of labeled data limits the development of medical VQA. In this paper, we propose a simple yet effective data augmentation method, VQAMix, to mitigate the data limitation problem. Specifically, VQAMix generates more labeled training samples by linearly combining a pair of VQA samples, which can be easily embedded into any visual-language model to boost performance. However, mixing two VQA samples would construct new connections between images and questions from different samples, which will cause the answers for those new fabricated image-question pairs to be missing or meaningless. To solve the missing answer problem, we first develop the Learning with Missing Labels (LML) strategy, which roughly excludes the missing answers. To alleviate the meaningless answer issue, we design the Learning with Conditional-mixed Labels (LCL) strategy, which further utilizes language-type prior to forcing the mixed pairs to have reasonable answers that belong to the same category. Experimental results on the VQA-RAD and PathVQA benchmarks show that our proposed method significantly improves the performance of the baseline by about 7% and 5% on the averaging result of two backbones, respectively. More importantly, VQAMix could improve confidence calibration and model interpretability, which is significant for medical VQA models in practical applications. All code and models are available at https://github.com/haifangong/VQAMix.

Evaluating Alarm Classifiers with High-confidence Data Programming

Classification of clinical alarms is at the heart of prioritization, suppression, integration, postponement, and other methods of mitigating alarm fatigue. Since these methods directly affect clinical care, alarm classifiers, such as intelligent suppression systems, need to be evaluated in terms of their sensitivity and specificity, which is typically calculated on a labeled dataset of alarms. Unfortunately, the collection and particularly labeling of such datasets requires substantial effort and time, thus deterring hospitals from investigating mitigations of alarm fatigue. This article develops a lightweight method for evaluating alarm classifiers without perfect alarm labels. The method relies on probabilistic labels obtained from data programming—a labeling paradigm based on combining noisy and cheap-to-obtain labeling heuristics. Based on these labels, the method produces confidence bounds for the sensitivity/specificity values from a hypothetical evaluation with manual labeling. Our experiments on five alarm datasets collected at Children’s Hospital of Philadelphia show that the proposed method provides accurate bounds on the classifier’s sensitivity/specificity, appropriately reflecting the uncertainty from noisy labeling and limited sample sizes.

Complementary features based prototype self-updating for few-shot learning

The goal of few-shot learning is to use limited labeled samples to complete independent classification tasks. The feature extractor of few-shot learning needs to have a stronger feature expression ability to generalize in unseen novel classes. To further enhance the expressive ability, in this paper, we propose an inherited feature extraction method, named Base and Meta Feature Extraction (BMFE). Base feature represents the task-irrelevant classification information of each sample. Meta feature obtained by the proposed Triplet Meta-train Mechanism (TMM) inherits the classification information and also contains the task-related meta information of each sample. We concatenate both the base and meta features to complementarily express the rich information of each sample. Besides, instead of relying on limited support samples to obtain the prototype, we propose a novel unsupervised prototype correction module, named Prototype Self-updating (PSU). All unlabeled query samples in a few-shot test task participate in the iterative updating of each prototype in the task without training. Extensive experiments prove that our overall method can obtain richer features by BMFE and more accurate prototypes by PSU. Our overall method outperforms state-of-the-art methods on miniImageNet and tiredImageNet datasets, and especially under the 1-shot case we obtains 78.45% and 81.21% classification accuracy respectively.

Cold-start active learning for image classification

Active learning (AL) aims to select valuable samples for labeling from an unlabeled sample pool to build a training dataset with minimal annotation cost. Traditional methods always require partially and initially labeled samples to start active selection and then query annotations of samples incrementally through several iterations. However, this scheme is not effective in the deep learning scenario. On the one hand, initially labeled sample sets are not always available in the beginning. On the other hand, the performance of the traditional model is usually poor in the early iterations due to limited training feedback. For the first time, we propose a cold-start AL model based on representative (CALR) sampling, which selects valuable samples without the need for an initial labeled set or the iterative feedback of the target models. Experiments on three image classification datasets, CIFAR-10, CIFAR-100 and Caltech-256, showed that CALR achieved a new state-of-the-art performance of AL in cold-start settings. Especially in low annotation budget conditions, our method can achieve up to a 10% performance increase compared to traditional methods. Furthermore, CALR can be combined with warm-start methods to improve the start-up efficiency while further breaking the performance ceiling of AL, which makes CALR have a broader application scenario.

Supervised and Semisupervised Manifold Embedded Knowledge Transfer in Motor Imagery-Based BCI.

A long calibration procedure limits the use in practice for a motor imagery (MI)-based brain-computer interface (BCI) system. To tackle this problem, we consider supervised and semisupervised transfer learning. However, it is a challenge for them to cope with high intersession/subject variability in the MI electroencephalographic (EEG) signals. Based on the framework of unsupervised manifold embedded knowledge transfer (MEKT), we propose a supervised MEKT algorithm (sMEKT) and a semisupervised MEKT algorithm (ssMEKT), respectively. sMEKT only has limited labelled samples from a target subject and abundant labelled samples from multiple source subjects. Compared to sMEKT, ssMEKT adds comparably more unlabelled samples from the target subject. After performing Riemannian alignment (RA) and tangent space mapping (TSM), both sMEKT and ssMEKT execute domain adaptation to shorten the differences among subjects. During domain adaptation, to make use of the available samples, two algorithms preserve the source domain discriminability, and ssMEKT preserves the geometric structure embedded in the labelled and unlabelled target domains. Moreover, to obtain a subject-specific classifier, sMEKT minimizes the joint probability distribution shift between the labelled target and source domains, whereas ssMEKT performs the joint probability distribution shift minimization between the unlabelled target domain and all labelled domains. Experimental results on two publicly available MI datasets demonstrate that our algorithms outperform the six competing algorithms, where the sizes of labelled and unlabelled target domains are variable. Especially for the target subjects with 10 labelled samples and 270/190 unlabelled samples, ssMEKT shows 5.27% and 2.69% increase in average accuracy on the two abovementioned datasets compared to the previous best semisupervised transfer learning algorithm (RA-regularized common spatial patterns-weighted adaptation regularization, RA-RCSP-wAR), respectively. Therefore, our algorithms can effectively reduce the need of labelled samples for the target subject, which is of importance for the MI-based BCI application.

Hybrid channel expansion and squeeze network for hyperspectral image classification

Deep learning has been widely used in hyperspectral image (HSI) classification tasks and has achieved good results in various fields. However, the large gap between a large number of parameters to be adjusted and the limited labeled samples of HSI will greatly increase the risk of model overfitting and inevitably consume a large amount of time and memory. A lightweight hybrid channel expansion and squeeze network is proposed to solve the small sample set problem. First, a channel expansion and squeeze module (CESM) based on depthwise convolution is designed to obtain higher fine-grained features by increasing the width of network channels. At the same time, the introduction of efficient channel attention module can emphasize the importance of each channel and effectively improve the representation of CESM compression output. Second, multidimensional CESM is used to reconstruct the entire feature extraction process and further reduce the model complexity while maintaining the intrinsic physical characteristics of HSI. Finally, the extracted high-level abstract features are imported into the global average pooling classifier to obtain the prediction probability of each category. Experimental results on Indian Pines, Pavia University, and Salinas Scene datasets show that the accuracy of the proposed method can reach 98.66%, 99.61%, and 99.84%, respectively, under limited labeled samples, and the computational cost is significantly lower than other advanced methods.

LIMU-BERT

Deep learning greatly empowers Inertial Measurement Unit (IMU) sensors for a wide range of sensing applications. Most existing works require substantial amounts of wellcurated labeled data to train IMU-based sensing models, which incurs high annotation and training costs. Compared with labeled data, unlabeled IMU data are abundant and easily accessible. This article presents a novel representation learning model that can make use of unlabeled IMU data and extract generalized rather than task-specific features. With the representations learned via our model, task-specific models trained with limited labeled samples can achieve superior performances in typical IMU sensing applications, such as Human Activity Recognition (HAR).

Attribute reduction algorithm of neighborhood rough set based on supervised granulation and its application

In neighborhood rough set theory, attribute reduction based on measure of information has important application significance. The influence of different decision classes was not considered for calculation of traditional conditional neighborhood entropy, and the improvement of algorithm based on conditional neighborhood entropy mainly includes of introducing multi granularity and different levels, while the mutual influence between samples with different labels is less considered. To solve this problem, this paper uses the supervised strategy to improve the conditional neighborhood entropy of three-layer granulation. By using two different neighborhood radii to adjust the mutual influence degree of different label samples, and by considering the mutual influence between conditional attributes through the feature complementary relationship, a neighborhood rough set attribute reduction algorithm based on supervised granulation is proposed. Experiment results on UCI data sets show that the proposed algorithm is superior to the traditional conditional neighborhood entropy algorithm in both aspects of reduction rate and reduction accuracy. Finally, the proposed algorithm is applied to the evaluation of fatigue life influencing factors of titanium alloy welded joints. The results of coupling relationship analysis show that the effect of joint type should be most seriously considered in the calculation of stress concentration factor. The results of influencing factors analysis show that the stress range has the highest weight among all the fatigue life influencing factors of titanium alloy welded joint.

Local Matrix Feature-Based Kernel Joint Sparse Representation for Hyperspectral Image Classification

Hyperspectral image (HSI) classification is one of the hot research topics in the field of remote sensing. The performance of HSI classification greatly depends on the effectiveness of feature learning or feature design. Traditional vector-based spectral–spatial features have shown good performance in HSI classification. However, when the number of labeled samples is limited, the performance of these vector-based features is degraded. To fully mine the discriminative features in small-sample case, a novel local matrix feature (LMF) was designed to reflect both the correlation between spectral pixels and the spectral bands in a local spatial neighborhood. In particular, the LMF is a linear combination of a local covariance matrix feature and a local correntropy matrix feature, where the former describes the correlation between spectral pixels and the latter measures the similarity between spectral bands. Based on the constructed LMFs, a simple Log-Euclidean distance-based linear kernel is introduced to measure the similarity between them, and an LMF-based kernel joint sparse representation (LMFKJSR) model is proposed for HSI classification. Due to the superior performance of region covariance and correntropy descriptors, the proposed LMFKJSR shows better results than existing vector-feature-based and matrix-feature-based support vector machine (SVM) and JSR methods on three well-known HSI data sets in the case of limited labeled samples.

Semisupervised Momentum Prototype Network for Gearbox Fault Diagnosis Under Limited Labeled Samples

It is difficult to obtain expensive labeled data in industrial fault diagnosis applications, which easily leads to overfitting of deep learning and restricts its extensive usage. Aiming at this issue, this article proposed an improved few-shot semisupervised learning method, called semisupervised momentum prototype network (SSMPN), to realize gearbox fault diagnosis under limited labeled samples. First, the proposed SSMPN utilizes the powerful few-shot learning ability of the prototype network to learn the feature mapping and obtains prototypes by using limited labeled samples. Then, a threshold selection based on Monte Carlo uncertainty is adopted in pseudo label learning to increase the confidence of pseudo labels. Finally, the expended labeled dataset is utilized to optimize feature extraction and the momentum prototype method is proposed to fine-tune the prototype of each category. The experiments on both test-bench and wind turbine gearbox fault diagnosis demonstrated that SSMPN is more effective than the comparable methods under the same situation.

GAN-based semi-supervised learning method for identification of the faulty feeder in resonant grounding distribution networks

Accurate identification of the faulty feeder after a single-phase-to-ground fault in an arc suppression coil grounding distribution network remains a significant challenge. Recently, intelligent learning-based methods and especially deep learning methods have attracted much attention. To our knowledge, these methods usually assume that the training samples are sufficient. However, it is difficult to obtain abundant labeled samples in a practical environment. In this paper, we propose a semi-supervised learning model based on generative adversarial networks (GAN) to reliably identify the faulty feeder. The discriminator of GAN classifies the synthetic data which is generated by the generator into a new class. We revise the loss functions of the discriminator and generator, and the discriminator and generator play a maximum and minimum game. The proposed method can significantly improve the performance of faulty feeder identification when a small percentage of training samples are labeled. Various types of faults are simulated to consist of the training dataset and testing dataset. The simulation test results, sensitive analysis, comparing analysis, and recorded data obtained in a full-scale test field test demonstrate the higher performance of the proposed method.

Visualizing the Conformational Dynamics of Membrane Receptors Using Single-Molecule FRET.

The ability of cells to respond to external signals is essential for cellular development, growth, and survival. To respond to a signal from the environment, a cell must be able to recognize and process it. This task mainly relies on the function of membrane receptors, whose role is to convert signals into the biochemical language of the cell. G protein-coupled receptors (GPCRs) constitute the largest family of membrane receptor proteins in humans. Among GPCRs, metabotropic glutamate receptors (mGluRs) are a unique subclass that function as obligate dimers and possess a large extracellular domain that contains the ligand-binding site. Recent advances in structural studies of mGluRs have improved the understanding of their activation process. However, the propagation of large-scale conformational changes through mGluRs during activation and modulation is poorly understood. Single-molecule fluorescence resonance energy transfer (smFRET) is a powerful technique to visualize and quantify the structural dynamics of biomolecules at the single-protein level. To visualize the dynamic process of mGluR2 activation, fluorescent conformational sensors based on unnatural amino acid (UAA) incorporation were developed that allowed site-specific protein labeling without perturbation of the native structure of receptors. The protocol described here explains how to perform these experiments, including the novel UAA labeling approach, sample preparation, and smFRET data acquisition and analysis. These strategies are generalizable and can be extended to investigate the conformational dynamics of a variety of membrane proteins.

Clustering-Guided Pairwise Metric Triplet Loss for Person Reidentification

Most of the loss functions proposed for person reidentification (Re-ID) are expected to be easy to deploy, efficiently improve network performance, and will not introduce redundant parameters. This study proposes a no-parameter and generic clustering-guided pairwise metric triplet (CPM-Triplet) loss based on the hard sample mining triplet loss for the metric learning loss. CPM-Triplet loss deploys two metrics: 1) the Euclidean metric and 2) the cosine metric, to complementarily improve the metric learning of the model. Paralleled to the Euclidean metric, the cosine metric quantifies the sample similarity in a different way to the Euclidean metric, which takes a different perspective to explore the distribution of samples. But the pairwise metric mainly improves the precision between dissimilar samples of the same label and could not solve the problem of excessive outliers. Therefore, the clustering-guided correction term was deployed to apply to all samples with the same label to mine the similarity in the samples, while weakening the influence of outliers in CPM-Triplet loss. Experiments conducted on four benchmark data sets show that the combination of the CPM-Triplet loss and the widely used Bag-of-Tricks baseline generally outperforms the baseline and numerous state-of-the-art methods studied in this article. The source code would be available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/weiyu-zeng/CPM-Triplet-loss</uri> .

Self-Supervised Transfer Learning Based on Domain Adaptation for Benign-Malignant Lung Nodule Classification on Thoracic CT.

The spatial heterogeneity is an important indicator of the malignancy of lung nodules in lung cancer diagnosis. Compared with 2D nodule CT images, the 3D volumes with entire nodule objects hold richer discriminative information. However, for deep learning methods driven by massive data, effectively capturing the 3D discriminative features of nodules in limited labeled samples is a challenging task. Different from previous models that proposed transfer learning models in a 2D pattern or learning from scratch 3D models, we develop a self-supervised transfer learning based on domain adaptation (SSTL-DA) 3D CNN framework for benign-malignant lung nodule classification. At first, a data pre-processing strategy termed adaptive slice selection (ASS) is developed to eliminate the redundant noise of the input samples with lung nodules. Then, the self-supervised learning network is constructed to learn robust image representations from CT images. Finally, a transfer learning method based on domain adaptation is designed to obtain discriminant features for classification. The proposed SSTL-DA method has been assessed on the LIDC-IDRI benchmark dataset, and it obtains an accuracy of 91.07% and an AUC of 95.84%. These results demonstrate that the SSTL-DA model achieves quite a competitive classification performance compared with some state-of-the-art approaches.

Semi-supervised medical image segmentation via uncertainty rectified pyramid consistency.

Despite that Convolutional Neural Networks (CNNs) have achieved promising performance in many medical image segmentation tasks, they rely on a large set of labeled images for training, which is expensive and time-consuming to acquire. Semi-supervised learning has shown the potential to alleviate this challenge by learning from a large set of unlabeled images and limited labeled samples. In this work, we present a simple yet efficient consistency regularization approach for semi-supervised medical image segmentation, called Uncertainty Rectified Pyramid Consistency (URPC). Inspired by the pyramid feature network, we chose a pyramid-prediction network that obtains a set of segmentation predictions at different scales. For semi-supervised learning, URPC learns from unlabeled data by minimizing the discrepancy between each of the pyramid predictions and their average. We further present multi-scale uncertainty rectification to boost the pyramid consistency regularization, where the rectification seeks to temper the consistency loss at outlier pixels that may have substantially different predictions than the average, potentially due to upsampling errors or lack of enough labeled data. Experiments on two public datasets and an in-house clinical dataset showed that: 1) URPC can achieve large performance improvement by utilizing unlabeled data and 2) Compared with five existing semi-supervised methods, URPC achieved better or comparable results with a simpler pipeline. Furthermore, we build a semi-supervised medical image segmentation codebase to boost research on this topic: https://github.com/HiLab-git/SSL4MIS.