Semi-supervised Learning Approach Research Articles

Permeability prediction of low-resolution porous media images using autoencoder-based convolutional neural network

Permeability prediction of porous media from numerical approaches is an important supplement for experimental measurements with the benefits of being more economical and efficient. However, the accuracy and reliability of traditional numerical approaches are strongly dependent on the high-resolution images of porous media, which greatly limits their implementation for engineering applications. Herein, a semi-supervised machine learning approach is proposed to predict the permeability of porous media from low-resolution images. This approach consists of an autoencoder (AE) module trained with unlabeled data to assist the backbone convolutional neural network (CNN) in the prediction by providing a mapping of the low-resolution porous media to high-resolution features. The low-resolution information from CNN trained with small amount of labeled data and high-resolution information from AE trained with larger amount of unlabeled data are comprehensively considered in this approach. The prediction performance of AE-CNN from low-resolution images is examined against the results from traditional approaches of CNN and lattice Boltzmann method (LBM) by the mean-square errors (MSE) and R-Squared (R2) calculations. Using 5-fold cross-validation method, the average value of R2 is 0.896 on the test dataset by AE-CNN, compared to 0.869 for the traditional CNN without the AE. The MSEs for AE-CNN are 0.022 and 0.064 on the training and test datasets respectively in the best-performance fold, while without AE, the MSEs for only CNN are 0.034 and 0.083 on the training and test datasets respectively in the best-performance fold, implying that AE modules can substantially improve the prediction performance from low-resolution images of porous media. As for the simulation results of LBM approach, its prediction reliability (average R2: 0.42; MSE: 0.37 and 0.36 in the best-performance fold) is extremely lower than that of CNN-based machine learning algorithms owing to huge numerical error at the blurred boundaries of low-resolution images.

Semi-supervised Transformer Network for Anomaly Detection in Cellular Internet of Things

Because of the lightning-fast expansion of the Internet of Things (IoT) technologies, an enormous amount of data has been produced. This traffic can be mined for information that can be used to identify and avoid intrusions into IoT networks. Despite the significant efforts that have been put into labeling Internet of Things traffic records, the total number of labeled records is still quite low, which makes it more difficult to detect intrusions. This study introduces a semi-supervised deep learning approach for intrusion detection (S2T-Net), in which we propose a temporal transformer module to empower the model to learn valuable interactions in cellular data. An improved spatial transformer is presented to capture local representation in the cellular traffic flow. At the same time, a multilevel semi-supervised training technique is used to account for the consecutive structure of the IoT traffic information. In order to provide effective real-time threat intelligence, the suggested S2T-Net can be tightly coupled into a cellular IoT network. Last but not least, empirical assessments on two current databases (CIC-IDS2017 and CIC-IDS2018) show that S2T-Net boosts intrusion detection accuracy and resilience while retaining resource-efficient computing.

A semi-supervised machine learning approach for in-process monitoring of laser powder bed fusion

Laser powder bed fusion (L-PBF), despite the tremendous potential in metal additive manufacturing, is still facing a significant barrier toward wider adoption due to the current lack of quality assurance. Notable efforts aiming at effective quality control of L-PBF products rely on using machine learning (ML) of monitoring data to either identify possible defects or predict the product quality. In this study, we propose a semi-supervised ML approach using layerwise monitoring images. We train the ML model using reference monitoring images to classify surface appearances of samples printed without defect and with a common type of defect in L-PBF, i.e., overheating. The trained ML model enables determination of overheated regions in L-PBF products during printing process. We then demonstrate our ML’s capability by performing prediction on a test sample having overhanging structures.

A semi-supervised clustering-based classification model for classifying imbalanced data streams in the presence of scarcely labelled data

Data streams are potentially infinite in length, fast changing and scarcely labelled. It is practically impossible to label all the observed instances. Online frameworks for classifying data streams are generally supervised in nature assuming the availability of labelled data and hence cannot be used for data streams. Semi-supervised learning (SSL) addresses this problem of scarcely labelled data by using large amount of unlabelled data together with labelled data to build classifiers. Data streams may also suffer from the problem of imbalanced data. Previous works in learning from data streams have analysed problems of imbalanced data. But to the best of our knowledge no work has applied semi-supervised learning approaches for classifying imbalanced data streams so far. This paper proposes a model using a semi-supervised clustering technique to classify an imbalanced data stream in the presence of scarcely labelled data. The results prove that the model outperforms many state-of-the-art techniques.

SemiSegSAR: A Semi-Supervised Segmentation Algorithm for Ship SAR Images

Automatic ship segmentation from high-resolution Synthetic Aperture Radar (SAR) remote sensing images has been a topic of interest that has gradually gained attention over the years due to the abundance of earth observation sensors. Recently, deep learning methods have provided a breakthrough increasing the performance greatly by using large amount of labeled data. Yet, the high cost related to the samples labeling and their scarcity result in significant limitation of their wide use. Therefore, it is crucial to overcome the unlabeled inputs challenge and develop semi-supervised learning approaches to enhance the machine learning models capacity. Our letter proposes a semi-supervised segmentation algorithm for SAR images named SemiSegSAR based on the use of Graph Signal Processing. This method includes instance segmentation; texture and statistical SAR features to represent the nodes of the graph; K-nearest neighbors to construct the graph; and Sobolev minimization algorithm to tackle the problem of semi-supervised semantic segmentation. The proposed algorithm is trained and tested using the publicly available SSDD and HRSID ship detection datasets. Experiments show that SemiSegSAR outperforms the current state-of-the-art semi-supervised and supervised methods while requiring only few labeled data.

A Novel Semi-Supervised Learning Approach for State of Health Monitoring of Maritime Lithium-Ion Batteries

A Novel Semi-Supervised Learning Approach for State of Health Monitoring of Maritime Lithium-Ion Batteries

Urban scene based Semantical Modulation for Pedestrian Detection

Despite recent progress, pedestrian detection still suffers from the troublesome problems of small objects, occlusions, and numerous false positives. Intuitively, the rich context information available from urban scenes could help determine the presence and location of pedestrians. For example, roads and sidewalks are good cues for potential pedestrians, while detections on buildings and trees are often false positives. However, most existing pedestrian detectors ignore or inadequately utilize semantic context. In this paper, in order to make full use of the urban-scene semantics to facilitate pedestrian detection, we propose a new method called Semantical Modulation based Pedestrian Detector (SMPD). First, for efficiency, a semantic prediction module is jointly learned with a baseline detector for semantic predictions. Second, a semantic integration module is designed to exploit the urban-scene semantic context for detection. Specifically, we force it to be an independent detection branch based solely on semantic information. In this way, together with the baseline detector, the fused detection results explicitly depend on both the learned appearance features and the scene context around pedestrians. In addition, while existing methods cannot be applied to the datasets where semantic annotations are not available for training, we introduce a semi-supervised transfer learning approach to make our method suitable for more scenarios. We demonstrate experimentally that, thanks to the integration of semantic context from urban scenes, SMPD can accurately detect small and occluded pedestrians, as well as effectively remove false positives. As a result, SMPD achieves the new state of the art on the Citypersons and Caltech datasets.

Semi-Supervised Learning for Anomaly Classification Using Partially Labeled Subsets

Abstract Machine learning and other data-driven methods have developed at a prolific rate for industrial applications due to the advent of industrial big data. However, industrial datasets may not be especially well-suited to supervised learning approaches that require extensive domain knowledge in the complete and accurate labeling of datasets. To address these challenges, a semi-supervised learning approach is proposed that makes use of partially labeled subsets. The proposed methodology is applied to high-dimensional in-process measurement data, utilizing a convolutional autoencoder (CAE) for unsupervised feature extraction. A multiclass extension for semi-supervised anomaly diagnosis is proposed that utilizes principal component analysis (PCA) as the basis for anomaly scoring, and the proposed approach intersects the results of targeted one-against-all phases on partially labeled sets to classify faults. Experiments in a case study on semiconductor manufacturing measurement data are performed to explore the relationship between latent features extracted and anomaly detection performance. The application of the proposed algorithm achieves a true positive detection rate of over 90% with false positive rate under 9% for both local and global anomaly types, with these results accomplished while reducing over 99% of the original input data dimensions. In addition, the approach also allows for positive samples to be identified that were previously undetected by human experts. These results are promising for the application of the proposed semi-supervised methodology in real industrial settings.

Low-rank supervised and semi-supervised multi-metric learning for classification

Multi-metric learning is an important technique for improving classification performance since learning a single metric is usually insufficient for complex data. Most of the existing multi-metric learning approaches have high computational complexity. In this work, two multi-metric learning frameworks proposed to perform supervised and semi-supervised classifications respectively. Based on the frameworks, we first design a low-rank multi-metric learning model (LSMML) for supervised classification, in which multiple local class metrics as well as one global metric are jointly trained. A joint regularization scheme, composed of LogDet divergence and low-rank term, is also well-designed to incorporate prior knowledge for improving generalization. By learning appropriate metrics, LSMML not only captures the local nonlinear discriminate information of each class to reduce the probability of misclassification but also enhances the stability, alleviates the computational burden, and avoids the risk of overfitting. Then, we extend LSMML to a semi-supervised learning scenario and propose a low-rank semi-supervised multi-metric learning approach (LSeMML) to process data with scarce labels. Alternating iterative algorithms are designed to optimize both LSMML and LSeMML. At each iteration, we only require to perform geodesic convex optimizations, with closed-form solutions and low computational cost. In supervised and semi-supervised settings respectively, numerical simulations are carried out on different databases, which shows that the proposed LSMML and LSeMML have a simple form, fast training speed, and good classification performance.

Predicting the hosts of prokaryotic viruses using GCN-based semi-supervised learning

BackgroundProkaryotic viruses, which infect bacteria and archaea, are the most abundant and diverse biological entities in the biosphere. To understand their regulatory roles in various ecosystems and to harness the potential of bacteriophages for use in therapy, more knowledge of viral-host relationships is required. High-throughput sequencing and its application to the microbiome have offered new opportunities for computational approaches for predicting which hosts particular viruses can infect. However, there are two main challenges for computational host prediction. First, the empirically known virus-host relationships are very limited. Second, although sequence similarity between viruses and their prokaryote hosts have been used as a major feature for host prediction, the alignment is either missing or ambiguous in many cases. Thus, there is still a need to improve the accuracy of host prediction.ResultsIn this work, we present a semi-supervised learning model, named HostG, to conduct host prediction for novel viruses. We construct a knowledge graph by utilizing both virus-virus protein similarity and virus-host DNA sequence similarity. Then graph convolutional network (GCN) is adopted to exploit viruses with or without known hosts in training to enhance the learning ability. During the GCN training, we minimize the expected calibrated error (ECE) to ensure the confidence of the predictions. We tested HostG on both simulated and real sequencing data and compared its performance with other state-of-the-art methods specifically designed for virus host classification (VHM-net, WIsH, PHP, HoPhage, RaFAH, vHULK, and VPF-Class).ConclusionHostG outperforms other popular methods, demonstrating the efficacy of using a GCN-based semi-supervised learning approach. A particular advantage of HostG is its ability to predict hosts from new taxa.

Rapid and flexible segmentation of electron microscopy data using few-shot machine learning

Automatic segmentation of key microstructural features in atomic-scale electron microscope images is critical to improved understanding of structure–property relationships in many important materials and chemical systems. However, the present paradigm involves time-intensive manual analysis that is inherently biased, error-prone, and unable to accommodate the large volumes of data produced by modern instrumentation. While more automated approaches have been proposed, many are not robust to a high variety of data, and do not generalize well to diverse microstructural features and material systems. Here, we present a flexible, semi-supervised few-shot machine learning approach for segmentation of scanning transmission electron microscopy images of three oxide material systems: (1) epitaxial heterostructures of SrTiO3/Ge, (2) La0.8Sr0.2FeO3 thin films, and (3) MoO3 nanoparticles. We demonstrate that the few-shot learning method is more robust against noise, more reconfigurable, and requires less data than conventional image analysis methods. This approach can enable rapid image classification and microstructural feature mapping needed for emerging high-throughput characterization and autonomous microscope platforms.

Potential diagnosis of COVID-19 from chest X-ray and CT findings using semi-supervised learning.

COVID-19 is an infectious disease, which has adversely affected public health and the economy across the world. On account of the highly infectious nature of the disease, rapid automated diagnosis of COVID-19 is urgently needed. A few recent findings suggest that chest X-rays and CT scans can be used by machine learning for the diagnosis of COVID-19. Herein, we employed semi-supervised learning (SSL) approaches to detect COVID-19 cases accurately by analyzing digital chest X-rays and CT scans. On a relatively small COVID-19 radiography dataset, which contains only 219 COVID-19 positive images, 1341 normal and 1345 viral pneumonia images, our algorithm, COVIDCon, which takes advantage of data augmentation, consistency regularization, and multicontrastive learning, attains 97.07% average class prediction accuracy, with 1000 labeled images, which is 7.65% better than the next best SSL method, virtual adversarial training. COVIDCon performs even better on a larger COVID-19 CT Scan dataset that contains 82,767 images. It achieved an excellent accuracy of 99.13%, at 20,000 labels, which is 6.45% better than the next best pseudo-labeling approach. COVIDCon outperforms other state-of-the-art algorithms at every label that we have investigated. These results demonstrate COVIDCon as the benchmark SSL algorithm for potential diagnosis of COVID-19 from chest X-rays and CT-Scans. Furthermore, COVIDCon performs exceptionally well in identifying COVID-19 positive cases from a completely unseen repository with a confirmed COVID-19 case history. COVIDCon, may provide a fast, accurate, and reliable method for screening COVID-19 patients.

NIMG-40. MRI-BASED ESTIMATION OF THE ABUNDANCE OF IMMUNOHISTOCHEMISTRY MARKERS IN GBM BRAIN USING DEEP LEARNING

Abstract Glioblastoma (GBM) is a devastating primary brain tumor known for its heterogeneity with a median survival of 15 months. Clinical imaging remains the primary modality to assess brain tumor response, but it is nearly impossible to distinguish between tumor growth and treatment response. Ki67 is a marker of active cell proliferation that shows inter- and intra-patient heterogeneity and should change under many therapies. In this work, we assessed the utility of a semi-supervised deep learning approach for regionally predicting high-vs-low Ki67 in GBM patients based on MRI. We used both labeled and unlabeled datasets to train the model. Labeled data included 114 MRI-localized biopsies from 43 unique GBM patients with available immunohistochemistry Ki67 labels. Unlabeled data included nine repeat routine pretreatment paired scans of newly-diagnosed GBM patients acquired within three days. Data augmentation techniques were utilized to enhance the size of our data and increase generalizability. Data was split between training, validation, and testing sets using 65-15-20 percent ratios. Model inputs were 16x16x3 patches around biopsies on T1Gd and T2 MRIs for labeled data, and around randomly selected patches inside the T2 abnormal region for unlabeled data. The network was a 4-conv layered VGG-inspired architecture. Training objective was accurate prediction of Ki67 in labeled patches and consistency in predictions across repeat unlabeled patches. We measured final model accuracy on held-out test samples. Our promising preliminary results suggest potential for deep learning in deconvolving the spatial heterogeneity of proliferative GBM subpopulations. If successful, this model can provide a non-invasive readout of cell proliferation and reveal the effectiveness of a given cytotoxic therapy dynamically during the patient's routine follow up. Further, the spatial resolution of our approach provides insights into the intra-tumoral heterogeneity of response which can be related to heterogeneity in localization of therapies (e.g. radiation therapy, drug dose heterogeneity).

Semi-Supervised Training for Positioning of Welding Seams.

Robotic welding often uses vision-based measurement to find the correct placement of the welding seam. Traditional machine vision methods work well in many cases but lack robustness when faced with variations in the manufacturing process or in the imaging conditions. While supervised deep neural networks have been successful in increasing accuracy and robustness in many real-world measurement applications, their success relies on labeled data. In this paper, we employ semi-supervised learning to simultaneously increase accuracy and robustness while avoiding expensive and time-consuming labeling efforts by a domain expert. While semi-supervised learning approaches for various image classification tasks exist, we purpose a novel algorithm for semi-supervised key-point detection for seam placement by a welding robot. We demonstrate that our approach can work robustly with as few as fifteen labeled images. In addition, our method utilizes full image resolution to enhance the accuracy of the key-point detection in seam placement.

Research of Metric Learning-Based Method for Person Re-Identification by Intelligent Computer Vision Technology

Person re-identification technology aims to establish an efficient metric model for similarity distance measurement of pedestrian images. Candidate images captured by different camera views are ranked according to their similarities to the target individual. However, the metric learning-based method, which is commonly used in similarity measurement, often failed in person re-identification tasks due to the drastic variations in appearance. The main reason for its low identification accuracy is that the metric learning method is over-fitting to the training data. Several types of metric learning methods which differ from each other by the distribution of sample pairs were summarized in this article for analysing and easing the metric learning methods’ over-fitting problem. Three different metric learning methods were tested on the VIPeR dataset. The distributions of the distance of the positive/negative training/test pairs are displayed to demonstrate the over-fitting problem. Then, a new metric model was proposed by combining the thoughts of binary classification and multi-class classification. Related verification experiments were conducted on VIPeR dataset. Besides, the semi-supervised metric learning approach was introduced to alleviate the over-fitting problem. The experimental results reflect gap between training pairs and test pairs in the metric subspace. Therefore, reducing the difference between training data and test data is a promising way to improve the identification accuracy of metric learning method.

Co-embedding: a semi-supervised multi-view representation learning approach

Co-embedding: a semi-supervised multi-view representation learning approach

A novel OC-SVM based ensemble learning framework for attack detection in AGC loop of power systems

This paper presents a Semi-supervised Learning approach for anomaly detection in the Automatic Generation Control loop of the power systems. The proposed technique is an ensemble method, which employs Gaussian One-class Support Vector Machine, K-Means clustering, and a novel Region Expanding Oversampling technique. Unlike previous research in this field, the proposed method learns from only healthy class data that are readily available in the historical database or can be obtained by offline simulation. The algorithm is proposed keeping in mind that the defenders may not predict the attack sequences ahead in time; hence, the attack instances are not utilised during the training. To validate the proposed method's performance in detecting unseen attacks, apart from general False Data Injection attacks, we have formulated two varieties of intelligently crafted attacks. At first, a coordinated stealthy attack is designed that has the potential to degrade system stability. Secondly, a new financially orientated market operation attack is formulated that induces higher financial settlements for the attacker-preferred generation companies. Finally, the performance of the proposed ensemble method is compared with existing one-class classifiers, which reveals the superiority of the proposed ensemble framework in terms of average geometric accuracy (98.79%), True Positive Rate (98.38%), and True Negative Rate (99.22%).

Semi-supervised semantic segmentation of prostate and organs-at-risk on 3D pelvic CT images

The recent development of deep learning approaches has revoluted medical data processing, including semantic segmentation, by dramatically improving performance. Automated segmentation can assist radiotherapy treatment planning by saving manual contouring efforts and reducing intra-observer and inter-observer variations. However, training effective deep learning models usually Requires a large amount of high-quality labeled data, often costly to collect. We developed a novel semi-supervised adversarial deep learning approach for 3D pelvic CT image semantic segmentation. Unlike supervised deep learning methods, the new approach can utilize both annotated and un-annotated data for training. It generates un-annotated synthetic data by a data augmentation scheme using generative adversarial networks (GANs). We applied the new approach to segmenting multiple organs in male pelvic CT images. CT images without annotations and GAN-synthesized un-annotated images were used in semi-supervised learning. Experimental results, evaluated by three metrics (Dice similarity coefficient, average Hausdorff distance, and average surface Hausdorff distance), showed that the new method achieved comparable performance with substantially fewer annotated images or better performance with the same amount of annotated data, outperforming the existing state-of-the-art methods.

Stratigraphic heterogeneity inference under limited borehole data and massive tunneling data during shield tunneling construction: a semi-supervised learning approach

It’s essential to identify the soil distribution for shield tunneling construction, especially in mixed ground conditions. Machine learning models have been conducted to infer the stratigraphic heterogeneity, but mostly relying on the geological survey report. The sparse borehole data has limited the identification accuracy of supervised learning models. Inspired by the perception that the shield tunneling data contains the geological information, a semi-supervised learning-based stratigraphic heterogeneity inference approach is presented in this study, which gets the utmost out of the limited borehole data and massive tunneling data. More specifically, a label propagation algorithm (LPA) soil identification model is developed. The rings with borehole data are defined as the labeled rings while other rings are unlabelled. The LPA is then employed to identify the soil distribution on the unlabelled rings with an iterative algorithm as the shield machine driving. A numerical experiment is conducted in Nanning metro line 1. The slurry pressured balanced shield was put forward in the mixed ground containing round gravel and mudstone. The field results show that the LPA approach can identify the mudstone distribution more accurately than traditional supervised learning methods. Input feature importance of tunneling data are also discussed

How and when to stop the co-training process

Co-training is a semi-supervised learning approach used when only a small set of the data that is available for training is labeled. By using multiple classifiers, the co-training process utilizes the small set of labeled data in order to label an additional set of samples. During this process, the classifiers gradually augment the training data in an iterative process in which a new co-training model is derived and used for labeling the unlabeled samples in each iteration. A few of the newly labeled samples are added in each iteration to the training dataset to improve the performance of the classifiers. The main challenge in applying co-training is to make sure that the co-trainer assigns accurate labels to the unlabeled samples. Many empirical studies showed that the performance (accuracy) of the co-trainer could not be further improved when a certain number of iterations was reached, and in some cases, the performance even declined if the process (i.e., labeling) continued. Despite this, no general solution has been suggested for identifying the optimal final co-training model or number of iterations before this decline. In this work, we propose a novel method aimed at selecting the near-optimal final co-training model among all models created in the various iterations according to a predefined measurement based solely on the unlabeled data. Experiments on nine open, publicly available and real-life datasets demonstrate that the proposed method outputs a near-optimal final co-training model compared to other co-training models created in the various iterations.