Automated classification of gas flow states in blast furnaces using top-camera imagery typically demands a large volume of labeled data,whose manual annotation is both labor-intensive and cost-prohibitive.To mitigate ...Automated classification of gas flow states in blast furnaces using top-camera imagery typically demands a large volume of labeled data,whose manual annotation is both labor-intensive and cost-prohibitive.To mitigate this challenge,we present an enhanced semi-supervised learning approach based on the Mean Teacher framework,incorporating a novel feature loss module to maximize classification performance with limited labeled samples.The model studies show that the proposed model surpasses both the baseline Mean Teacher model and fully supervised method in accuracy.Specifically,for datasets with 20%,30%,and 40%label ratios,using a single training iteration,the model yields accuracies of 78.61%,82.21%,and 85.2%,respectively,while multiple-cycle training iterations achieves 82.09%,81.97%,and 81.59%,respectively.Furthermore,scenario-specific training schemes are introduced to support diverse deployment need.These findings highlight the potential of the proposed technique in minimizing labeling requirements and advancing intelligent blast furnace diagnostics.展开更多
Background:Stomach cancer(SC)is one of the most lethal malignancies worldwide due to late-stage diagnosis and limited treatment.The transcriptomic,epigenomic,and proteomic,etc.,omics datasets generated by high-through...Background:Stomach cancer(SC)is one of the most lethal malignancies worldwide due to late-stage diagnosis and limited treatment.The transcriptomic,epigenomic,and proteomic,etc.,omics datasets generated by high-throughput sequencing technology have become prominent in biomedical research,and they reveal molecular aspects of cancer diagnosis and therapy.Despite the development of advanced sequencing technology,the presence of high-dimensionality in multi-omics data makes it challenging to interpret the data.Methods:In this study,we introduce RankXLAN,an explainable ensemble-based multi-omics framework that integrates feature selection(FS),ensemble learning,bioinformatics,and in-silico validation for robust biomarker detection,potential therapeutic drug-repurposing candidates’identification,and classification of SC.To enhance the interpretability of the model,we incorporated explainable artificial intelligence(SHapley Additive exPlanations analysis),as well as accuracy,precision,F1-score,recall,cross-validation,specificity,likelihood ratio(LR)+,LR−,and Youden index results.Results:The experimental results showed that the top four FS algorithms achieved improved results when applied to the ensemble learning classification model.The proposed ensemble model produced an area under the curve(AUC)score of 0.994 for gene expression,0.97 for methylation,and 0.96 for miRNA expression data.Through the integration of bioinformatics and ML approach of the transcriptomic and epigenomic multi-omics dataset,we identified potential marker genes,namely,UBE2D2,HPCAL4,IGHA1,DPT,and FN3K.In-silico molecular docking revealed a strong binding affinity between ANKRD13C and the FDA-approved drug Everolimus(binding affinity−10.1 kcal/mol),identifying ANKRD13C as a potential therapeutic drug-repurposing target for SC.Conclusion:The proposed framework RankXLAN outperforms other existing frameworks for serum biomarker identification,therapeutic target identification,and SC classification with multi-omics datasets.展开更多
Accurate prediction of the remaining useful life(RUL)is crucial for the design and management of lithium-ion batteries.Although various machine learning models offer promising predictions,one critical but often overlo...Accurate prediction of the remaining useful life(RUL)is crucial for the design and management of lithium-ion batteries.Although various machine learning models offer promising predictions,one critical but often overlooked challenge is their demand for considerable run-to-failure data for training.Collection of such training data leads to prohibitive testing efforts as the run-to-failure tests can last for years.Here,we propose a semi-supervised representation learning method to enhance prediction accuracy by learning from data without RUL labels.Our approach builds on a sophisticated deep neural network that comprises an encoder and three decoder heads to extract time-dependent representation features from short-term battery operating data regardless of the existence of RUL labels.The approach is validated using three datasets collected from 34 batteries operating under various conditions,encompassing over 19,900 charge and discharge cycles.Our method achieves a root mean squared error(RMSE)within 25 cycles,even when only 1/50 of the training dataset is labelled,representing a reduction of 48%compared to the conventional approach.We also demonstrate the method's robustness with varying numbers of labelled data and different weights assigned to the three decoder heads.The projection of extracted features in low space reveals that our method effectively learns degradation features from unlabelled data.Our approach highlights the promise of utilising semi-supervised learning to reduce the data demand for reliability monitoring of energy devices.展开更多
Deep Learning(DL)is such a powerful tool that we have seen tremendous success in areas such as Computer Vision,Speech Recognition,and Natural Language Processing.Since Automated Modulation Classification(AMC)is an imp...Deep Learning(DL)is such a powerful tool that we have seen tremendous success in areas such as Computer Vision,Speech Recognition,and Natural Language Processing.Since Automated Modulation Classification(AMC)is an important part in Cognitive Radio Networks,we try to explore its potential in solving signal modulation recognition problem.It cannot be overlooked that DL model is a complex model,thus making them prone to over-fitting.DL model requires many training data to combat with over-fitting,but adding high quality labels to training data manually is not always cheap and accessible,especially in real-time system,which may counter unprecedented data in dataset.Semi-supervised Learning is a way to exploit unlabeled data effectively to reduce over-fitting in DL.In this paper,we extend Generative Adversarial Networks(GANs)to the semi-supervised learning will show it is a method can be used to create a more dataefficient classifier.展开更多
Through semi-supervised learning and knowledge inheritance,a novel Takagi-Sugeno-Kang(TSK)fuzzy system framework is proposed for epilepsy data classification in this study.The new method is based on the maximum mean d...Through semi-supervised learning and knowledge inheritance,a novel Takagi-Sugeno-Kang(TSK)fuzzy system framework is proposed for epilepsy data classification in this study.The new method is based on the maximum mean discrepancy(MMD)method and TSK fuzzy system,as a basic model for the classification of epilepsy data.First,formedical data,the interpretability of TSK fuzzy systems can ensure that the prediction results are traceable and safe.Second,in view of the deviation in the data distribution between the real source domain and the target domain,MMD is used to measure the distance between different data distributions.The objective function is constructed according to the MMD distance,and the distribution distance of different datasets is minimized to find the similar characteristics of different datasets.We introduce semi-supervised learning to further explore the relationship between data.Based on the MMD method,a semi-supervised learning(SSL)-MMD method is constructed by using pseudo-tags to realize the data distribution alignment of the same category.In addition,the idea of knowledge dissemination is used to learn pseudo-tags as additional data features.Finally,for epilepsy classification,the cross-domain TSK fuzzy system uses the cross-entropy function as the objective function and adopts the back-propagation strategy to optimize the parameters.The experimental results show that the new method can process complex epilepsy data and identify whether patients have epilepsy.展开更多
Malaria is a lethal disease responsible for thousands of deaths worldwide every year.Manual methods of malaria diagnosis are timeconsuming that require a great deal of human expertise and efforts.Computerbased automat...Malaria is a lethal disease responsible for thousands of deaths worldwide every year.Manual methods of malaria diagnosis are timeconsuming that require a great deal of human expertise and efforts.Computerbased automated diagnosis of diseases is progressively becoming popular.Although deep learning models show high performance in the medical field,it demands a large volume of data for training which is hard to acquire for medical problems.Similarly,labeling of medical images can be done with the help of medical experts only.Several recent studies have utilized deep learning models to develop efficient malaria diagnostic system,which showed promising results.However,the most common problem with these models is that they need a large amount of data for training.This paper presents a computer-aided malaria diagnosis system that combines a semi-supervised generative adversarial network and transfer learning.The proposed model is trained in a semi-supervised manner and requires less training data than conventional deep learning models.Performance of the proposed model is evaluated on a publicly available dataset of blood smear images(with malariainfected and normal class)and achieved a classification accuracy of 96.6%.展开更多
In aerospace industry,gears are the most common parts of a mechanical transmission system.Gear pitting faults could cause the transmission system to crash and give rise to safety disaster.It is always a challenging pr...In aerospace industry,gears are the most common parts of a mechanical transmission system.Gear pitting faults could cause the transmission system to crash and give rise to safety disaster.It is always a challenging problem to diagnose the gear pitting condition directly through the raw signal of vibration.In this paper,a novel method named augmented deep sparse autoencoder(ADSAE)is proposed.The method can be used to diagnose the gear pitting fault with relatively few raw vibration signal data.This method is mainly based on the theory of pitting fault diagnosis and creatively combines with both data augmentation ideology and the deep sparse autoencoder algorithm for the fault diagnosis of gear wear.The effectiveness of the proposed method is validated by experiments of six types of gear pitting conditions.The results show that the ADSAE method can effectively increase the network generalization ability and robustness with very high accuracy.This method can effectively diagnose different gear pitting conditions and show the obvious trend according to the severity of gear wear faults.The results obtained by the ADSAE method proposed in this paper are compared with those obtained by other common deep learning methods.This paper provides an important insight into the field of gear fault diagnosis based on deep learning and has a potential practical application value.展开更多
Large amounts of labeled data are usually needed for training deep neural networks in medical image studies,particularly in medical image classification.However,in the field of semi-supervised medical image analysis,l...Large amounts of labeled data are usually needed for training deep neural networks in medical image studies,particularly in medical image classification.However,in the field of semi-supervised medical image analysis,labeled data is very scarce due to patient privacy concerns.For researchers,obtaining high-quality labeled images is exceedingly challenging because it involves manual annotation and clinical understanding.In addition,skin datasets are highly suitable for medical image classification studies due to the inter-class relationships and the inter-class similarities of skin lesions.In this paper,we propose a model called Coalition Sample Relation Consistency(CSRC),a consistency-based method that leverages Canonical Correlation Analysis(CCA)to capture the intrinsic relationships between samples.Considering that traditional consistency-based models only focus on the consistency of prediction,we additionally explore the similarity between features by using CCA.We enforce feature relation consistency based on traditional models,encouraging the model to learn more meaningful information from unlabeled data.Finally,considering that cross-entropy loss is not as suitable as the supervised loss when studying with imbalanced datasets(i.e.,ISIC 2017 and ISIC 2018),we improve the supervised loss to achieve better classification accuracy.Our study shows that this model performs better than many semi-supervised methods.展开更多
Machine learning techniques and a dataset of five wells from the Rawat oilfield in Sudan containing 93,925 samples per feature(seven well logs and one facies log) were used to classify four facies. Data preprocessing ...Machine learning techniques and a dataset of five wells from the Rawat oilfield in Sudan containing 93,925 samples per feature(seven well logs and one facies log) were used to classify four facies. Data preprocessing and preparation involve two processes: data cleaning and feature scaling. Several machine learning algorithms, including Linear Regression(LR), Decision Tree(DT), Support Vector Machine(SVM),Random Forest(RF), and Gradient Boosting(GB) for classification, were tested using different iterations and various combinations of features and parameters. The support vector radial kernel training model achieved an accuracy of 72.49% without grid search and 64.02% with grid search, while the blind-well test scores were 71.01% and 69.67%, respectively. The Decision Tree(DT) Hyperparameter Optimization model showed an accuracy of 64.15% for training and 67.45% for testing. In comparison, the Decision Tree coupled with grid search yielded better results, with a training score of 69.91% and a testing score of67.89%. The model's validation was carried out using the blind well validation approach, which achieved an accuracy of 69.81%. Three algorithms were used to generate the gradient-boosting model. During training, the Gradient Boosting classifier achieved an accuracy score of 71.57%, and during testing, it achieved 69.89%. The Grid Search model achieved a higher accuracy score of 72.14% during testing. The Extreme Gradient Boosting model had the lowest accuracy score, with only 66.13% for training and66.12% for testing. For validation, the Gradient Boosting(GB) classifier model achieved an accuracy score of 75.41% on the blind well test, while the Gradient Boosting with Grid Search achieved an accuracy score of 71.36%. The Enhanced Random Forest and Random Forest with Bagging algorithms were the most effective, with validation accuracies of 78.30% and 79.18%, respectively. However, the Random Forest and Random Forest with Grid Search models displayed significant variance between their training and testing scores, indicating the potential for overfitting. Random Forest(RF) and Gradient Boosting(GB) are highly effective for facies classification because they handle complex relationships and provide high predictive accuracy. The choice between the two depends on specific project requirements, including interpretability, computational resources, and data nature.展开更多
It is crucial to maintain the safe and stable operation of distribution transformers,which constitute a key part of power systems.In the event of transformer failure,the fault type must be diagnosed in a timely and ac...It is crucial to maintain the safe and stable operation of distribution transformers,which constitute a key part of power systems.In the event of transformer failure,the fault type must be diagnosed in a timely and accurate manner.To this end,a transformer fault diagnosis method based on infrared image processing and semi-supervised learning is proposed herein.First,we perform feature extraction on the collected infrared-image data to extract temperature,texture,and shape features as the model reference vectors.Then,a generative adversarial network(GAN)is constructed to generate synthetic samples for the minority subset of labelled samples.The proposed method can learn information from unlabeled sample data,unlike conventional supervised learning methods.Subsequently,a semi-supervised graph model is trained on the entire dataset,i.e.,both labeled and unlabeled data.Finally,we test the proposed model on an actual dataset collected from a Chinese electricity provider.The experimental results show that the use of feature extraction,sample generation,and semi-supervised learning model can improve the accuracy of transformer fault classification.This verifies the effectiveness of the proposed method.展开更多
The majority of big data analytics applied to transportation datasets suffer from being too domain-specific,that is,they draw conclusions for a dataset based on analytics on the same dataset.This makes models trained ...The majority of big data analytics applied to transportation datasets suffer from being too domain-specific,that is,they draw conclusions for a dataset based on analytics on the same dataset.This makes models trained from one domain(e.g.taxi data)applies badly to a different domain(e.g.Uber data).To achieve accurate analyses on a new domain,substantial amounts of data must be available,which limits practical applications.To remedy this,we propose to use semi-supervised and active learning of big data to accomplish the domain adaptation task:Selectively choosing a small amount of datapoints from a new domain while achieving comparable performances to using all the datapoints.We choose the New York City(NYC)transportation data of taxi and Uber as our dataset,simulating different domains with 90%as the source data domain for training and the remaining 10%as the target data domain for evaluation.We propose semi-supervised and active learning strategies and apply it to the source domain for selecting datapoints.Experimental results show that our adaptation achieves a comparable performance of using all datapoints while using only a fraction of them,substantially reducing the amount of data required.Our approach has two major advantages:It can make accurate analytics and predictions when big datasets are not available,and even if big datasets are available,our approach chooses the most informative datapoints out of the dataset,making the process much more efficient without having to process huge amounts of data.展开更多
Intelligent seismic facies identification based on deep learning can alleviate the time-consuming and labor-intensive problem of manual interpretation,which has been widely applied.Supervised learning can realize faci...Intelligent seismic facies identification based on deep learning can alleviate the time-consuming and labor-intensive problem of manual interpretation,which has been widely applied.Supervised learning can realize facies identification with high efficiency and accuracy;however,it depends on the usage of a large amount of well-labeled data.To solve this issue,we propose herein an incremental semi-supervised method for intelligent facies identification.Our method considers the continuity of the lateral variation of strata and uses cosine similarity to quantify the similarity of the seismic data feature domain.The maximum-diff erence sample in the neighborhood of the currently used training data is then found to reasonably expand the training sets.This process continuously increases the amount of training data and learns its distribution.We integrate old knowledge while absorbing new ones to realize incremental semi-supervised learning and achieve the purpose of evolving the network models.In this work,accuracy and confusion matrix are employed to jointly control the predicted results of the model from both overall and partial aspects.The obtained values are then applied to a three-dimensional(3D)real dataset and used to quantitatively evaluate the results.Using unlabeled data,our proposed method acquires more accurate and stable testing results compared to conventional supervised learning algorithms that only use well-labeled data.A considerable improvement for small-sample categories is also observed.Using less than 1%of the training data,the proposed method can achieve an average accuracy of over 95%on the 3D dataset.In contrast,the conventional supervised learning algorithm achieved only approximately 85%.展开更多
Direct online measurement on product quality of industrial processes is difficult to be realized,which leads to a large number of unlabeled samples in modeling data.Therefore,it needs to employ semi-supervised learnin...Direct online measurement on product quality of industrial processes is difficult to be realized,which leads to a large number of unlabeled samples in modeling data.Therefore,it needs to employ semi-supervised learning(SSL)method to establish the soft sensor model of product quality.Considering the slow time-varying characteristic of industrial processes,the model parameters should be updated smoothly.According to this characteristic,this paper proposes an online adaptive semi-supervised learning algorithm based on random vector functional link network(RVFLN),denoted as OAS-RVFLN.By introducing a L2-fusion term that can be seen a weight deviation constraint,the proposed algorithm unifies the offline and online learning,and achieves smoothness of model parameter update.Empirical evaluations both on benchmark testing functions and datasets reveal that the proposed OAS-RVFLN can outperform the conventional methods in learning speed and accuracy.Finally,the OAS-RVFLN is applied to the coal dense medium separation process in coal industry to estimate the ash content of coal product,which further verifies its effectiveness and potential of industrial application.展开更多
Intrusion detection involves identifying unauthorized network activity and recognizing whether the data constitute an abnormal network transmission.Recent research has focused on using semi-supervised learning mechani...Intrusion detection involves identifying unauthorized network activity and recognizing whether the data constitute an abnormal network transmission.Recent research has focused on using semi-supervised learning mechanisms to identify abnormal network traffic to deal with labeled and unlabeled data in the industry.However,real-time training and classifying network traffic pose challenges,as they can lead to the degradation of the overall dataset and difficulties preventing attacks.Additionally,existing semi-supervised learning research might need to analyze the experimental results comprehensively.This paper proposes XA-GANomaly,a novel technique for explainable adaptive semi-supervised learning using GANomaly,an image anomalous detection model that dynamically trains small subsets to these issues.First,this research introduces a deep neural network(DNN)-based GANomaly for semi-supervised learning.Second,this paper presents the proposed adaptive algorithm for the DNN-based GANomaly,which is validated with four subsets of the adaptive dataset.Finally,this study demonstrates a monitoring system that incorporates three explainable techniques—Shapley additive explanations,reconstruction error visualization,and t-distributed stochastic neighbor embedding—to respond effectively to attacks on traffic data at each feature engineering stage,semi-supervised learning,and adaptive learning.Compared to other single-class classification techniques,the proposed DNN-based GANomaly achieves higher scores for Network Security Laboratory-Knowledge Discovery in Databases and UNSW-NB15 datasets at 13%and 8%of F1 scores and 4.17%and 11.51%for accuracy,respectively.Furthermore,experiments of the proposed adaptive learning reveal mostly improved results over the initial values.An analysis and monitoring system based on the combination of the three explainable methodologies is also described.Thus,the proposed method has the potential advantages to be applied in practical industry,and future research will explore handling unbalanced real-time datasets in various scenarios.展开更多
In the upcoming large-scale Internet of Things(Io T),it is increasingly challenging to defend against malicious traffic,due to the heterogeneity of Io T devices and the diversity of Io T communication protocols.In thi...In the upcoming large-scale Internet of Things(Io T),it is increasingly challenging to defend against malicious traffic,due to the heterogeneity of Io T devices and the diversity of Io T communication protocols.In this paper,we propose a semi-supervised learning-based approach to detect malicious traffic at the access side.It overcomes the resource-bottleneck problem of traditional malicious traffic defenders which are deployed at the victim side,and also is free of labeled traffic data in model training.Specifically,we design a coarse-grained behavior model of Io T devices by self-supervised learning with unlabeled traffic data.Then,we fine-tune this model to improve its accuracy in malicious traffic detection by adopting a transfer learning method using a small amount of labeled data.Experimental results show that our method can achieve the accuracy of 99.52%and the F1-score of 99.52%with only 1%of the labeled training data based on the CICDDoS2019 dataset.Moreover,our method outperforms the stateof-the-art supervised learning-based methods in terms of accuracy,precision,recall and F1-score with 1%of the training data.展开更多
Radio frequency fingerprinting(RFF)is a remarkable lightweight authentication scheme to support rapid and scalable identification in the internet of things(IoT)systems.Deep learning(DL)is a critical enabler of RFF ide...Radio frequency fingerprinting(RFF)is a remarkable lightweight authentication scheme to support rapid and scalable identification in the internet of things(IoT)systems.Deep learning(DL)is a critical enabler of RFF identification by leveraging the hardware-level features.However,traditional supervised learning methods require huge labeled training samples.Therefore,how to establish a highperformance supervised learning model with few labels under practical application is still challenging.To address this issue,we in this paper propose a novel RFF semi-supervised learning(RFFSSL)model which can obtain a better performance with few meta labels.Specifically,the proposed RFFSSL model is constituted by a teacher-student network,in which the student network learns from the pseudo label predicted by the teacher.Then,the output of the student model will be exploited to improve the performance of teacher among the labeled data.Furthermore,a comprehensive evaluation on the accuracy is conducted.We derive about 50 GB real long-term evolution(LTE)mobile phone’s raw signal datasets,which is used to evaluate various models.Experimental results demonstrate that the proposed RFFSSL scheme can achieve up to 97%experimental testing accuracy over a noisy environment only with 10%labeled samples when training samples equal to 2700.展开更多
The rapid development of the Internet brings a variety of original information including text information, audio information, etc. However, it is difficult to find the most useful knowledge rapidly and accurately beca...The rapid development of the Internet brings a variety of original information including text information, audio information, etc. However, it is difficult to find the most useful knowledge rapidly and accurately because of its huge number. Automatic text classification technology based on machine learning can classify a large number of natural language documents into the corresponding subject categories according to its correct semantics. It is helpful to grasp the text information directly. By learning from a set of hand-labeled documents,we obtain the traditional supervised classifier for text categorization(TC). However, labeling all data by human is labor intensive and time consuming. To solve this problem, some scholars proposed a semi-supervised learning method to train classifier, but it is unfeasible for various kinds and great number of Web data since it still needs a part of hand-labeled data. In 2012, Li et al. invented a fully automatic categorization approach for text(FACT)based on supervised learning, where no manual labeling efforts are required. But automatically labeling all data can bring noise into experiment and cause the fact that the result cannot meet the accuracy requirement. We put forward a new idea that part of data with high accuracy can be automatically tagged based on the semantic of category name, then a semi-supervised way is taken to train classifier with both labeled and unlabeled data,and ultimately a precise classification of massive text data can be achieved. The empirical experiments show that the method outperforms the supervised support vector machine(SVM) in terms of both F1 performance and classification accuracy in most cases. It proves the effectiveness of the semi-supervised algorithm in automatic TC.展开更多
Artificial intelligent based dialog systems are getting attention from both business and academic communities.The key parts for such intelligent chatbot systems are domain classification,intent detection,and named ent...Artificial intelligent based dialog systems are getting attention from both business and academic communities.The key parts for such intelligent chatbot systems are domain classification,intent detection,and named entity recognition.Various supervised,unsupervised,and hybrid approaches are used to detect each field.Such intelligent systems,also called natural language understanding systems analyze user requests in sequential order:domain classification,intent,and entity recognition based on the semantic rules of the classified domain.This sequential approach propagates the downstream error;i.e.,if the domain classification model fails to classify the domain,intent and entity recognition fail.Furthermore,training such intelligent system necessitates a large number of user-annotated datasets for each domain.This study proposes a single joint predictive deep neural network framework based on long short-term memory using only a small user-annotated dataset to address these issues.It investigates value added by incorporating unlabeled data from user chatting logs into multi-domain spoken language understanding systems.Systematic experimental analysis of the proposed joint frameworks,along with the semi-supervised multi-domain model,using open-source annotated and unannotated utterances shows robust improvement in the predictive performance of the proposed multi-domain intelligent chatbot over a base joint model and joint model based on adversarial learning.展开更多
For the classification problem in practice,one of the challenging issues is to obtain enough labeled data for training.Moreover,even if such labeled data has been sufficiently accumulated,most datasets often exhibit l...For the classification problem in practice,one of the challenging issues is to obtain enough labeled data for training.Moreover,even if such labeled data has been sufficiently accumulated,most datasets often exhibit long-tailed distribution with heavy class imbalance,which results in a biased model towards a majority class.To alleviate such class imbalance,semisupervised learning methods using additional unlabeled data have been considered.However,as a matter of course,the accuracy is much lower than that from supervised learning.In this study,under the assumption that additional unlabeled data is available,we propose the iterative semi-supervised learning algorithms,which iteratively correct the labeling of the extra unlabeled data based on softmax probabilities.The results show that the proposed algorithms provide the accuracy as high as that from the supervised learning.To validate the proposed algorithms,we tested on the two scenarios:with the balanced unlabeled dataset and with the imbalanced unlabeled dataset.Under both scenarios,our proposed semi-supervised learning algorithms provided higher accuracy than previous state-of-the-arts.Code is available at https://github.com/HeewonChung92/iterative-semi-learning.展开更多
Significant breakthroughs in medical image registration have been achieved using deep neural networks(DNNs).However,DNN-based end-to-end registration methods often require large quantities of data or adequate annotati...Significant breakthroughs in medical image registration have been achieved using deep neural networks(DNNs).However,DNN-based end-to-end registration methods often require large quantities of data or adequate annotations for training.To leverage the intensity information of abundant unlabeled images,unsupervised registration methods commonly employ intensity-based similarity measures to optimize the network parameters.However,finding a sufficiently robust measure can be challenging for specific registration applications.Weakly supervised registration methods use anatomical labels to estimate the deformation between images.High-level structural information in label images is more reliable and practical for estimating the voxel correspondence of anatomic regions of interest between images,whereas label images are extremely difficult to collect.In this paper,we propose a two-stage semi-supervised learning framework for medical image registration,which consists of unsupervised and weakly supervised registration networks.The proposed semi-supervised learning framework is trained with intensity information from available images,label information from a relatively small number of labeled images and pseudo-label information from unlabeled images.Experimental results on two datasets(cardiac and abdominal images)demonstrate the efficacy and efficiency of this method in intra-and inter-modality medical image registrations,as well as its superior performance when a vast amount of unlabeled data and a small set of annotations are available.Our code is publicly available at at https://github.com/jdq818/SeRN.展开更多
基金financial support provided by the Natural Science Foundation of Hebei Province,China(No.E2024105036)the Tangshan Talent Funding Project,China(Nos.B202302007 and A2021110015)+1 种基金the National Natural Science Foundation of China(No.52264042)the Australian Research Council(No.IH230100010)。
文摘Automated classification of gas flow states in blast furnaces using top-camera imagery typically demands a large volume of labeled data,whose manual annotation is both labor-intensive and cost-prohibitive.To mitigate this challenge,we present an enhanced semi-supervised learning approach based on the Mean Teacher framework,incorporating a novel feature loss module to maximize classification performance with limited labeled samples.The model studies show that the proposed model surpasses both the baseline Mean Teacher model and fully supervised method in accuracy.Specifically,for datasets with 20%,30%,and 40%label ratios,using a single training iteration,the model yields accuracies of 78.61%,82.21%,and 85.2%,respectively,while multiple-cycle training iterations achieves 82.09%,81.97%,and 81.59%,respectively.Furthermore,scenario-specific training schemes are introduced to support diverse deployment need.These findings highlight the potential of the proposed technique in minimizing labeling requirements and advancing intelligent blast furnace diagnostics.
基金the Deanship of Research and Graduate Studies at King Khalid University,KSA,for funding this work through the Large Research Project under grant number RGP2/164/46.
文摘Background:Stomach cancer(SC)is one of the most lethal malignancies worldwide due to late-stage diagnosis and limited treatment.The transcriptomic,epigenomic,and proteomic,etc.,omics datasets generated by high-throughput sequencing technology have become prominent in biomedical research,and they reveal molecular aspects of cancer diagnosis and therapy.Despite the development of advanced sequencing technology,the presence of high-dimensionality in multi-omics data makes it challenging to interpret the data.Methods:In this study,we introduce RankXLAN,an explainable ensemble-based multi-omics framework that integrates feature selection(FS),ensemble learning,bioinformatics,and in-silico validation for robust biomarker detection,potential therapeutic drug-repurposing candidates’identification,and classification of SC.To enhance the interpretability of the model,we incorporated explainable artificial intelligence(SHapley Additive exPlanations analysis),as well as accuracy,precision,F1-score,recall,cross-validation,specificity,likelihood ratio(LR)+,LR−,and Youden index results.Results:The experimental results showed that the top four FS algorithms achieved improved results when applied to the ensemble learning classification model.The proposed ensemble model produced an area under the curve(AUC)score of 0.994 for gene expression,0.97 for methylation,and 0.96 for miRNA expression data.Through the integration of bioinformatics and ML approach of the transcriptomic and epigenomic multi-omics dataset,we identified potential marker genes,namely,UBE2D2,HPCAL4,IGHA1,DPT,and FN3K.In-silico molecular docking revealed a strong binding affinity between ANKRD13C and the FDA-approved drug Everolimus(binding affinity−10.1 kcal/mol),identifying ANKRD13C as a potential therapeutic drug-repurposing target for SC.Conclusion:The proposed framework RankXLAN outperforms other existing frameworks for serum biomarker identification,therapeutic target identification,and SC classification with multi-omics datasets.
基金supported by the National Natural Science Foundation of China(No.52207229)the Key Research and Development Program of Ningxia Hui Autonomous Region of China(No.2024BEE02003)+1 种基金the financial support from the AEGiS Research Grant 2024,University of Wollongong(No.R6254)the financial support from the China Scholarship Council(No.202207550010).
文摘Accurate prediction of the remaining useful life(RUL)is crucial for the design and management of lithium-ion batteries.Although various machine learning models offer promising predictions,one critical but often overlooked challenge is their demand for considerable run-to-failure data for training.Collection of such training data leads to prohibitive testing efforts as the run-to-failure tests can last for years.Here,we propose a semi-supervised representation learning method to enhance prediction accuracy by learning from data without RUL labels.Our approach builds on a sophisticated deep neural network that comprises an encoder and three decoder heads to extract time-dependent representation features from short-term battery operating data regardless of the existence of RUL labels.The approach is validated using three datasets collected from 34 batteries operating under various conditions,encompassing over 19,900 charge and discharge cycles.Our method achieves a root mean squared error(RMSE)within 25 cycles,even when only 1/50 of the training dataset is labelled,representing a reduction of 48%compared to the conventional approach.We also demonstrate the method's robustness with varying numbers of labelled data and different weights assigned to the three decoder heads.The projection of extracted features in low space reveals that our method effectively learns degradation features from unlabelled data.Our approach highlights the promise of utilising semi-supervised learning to reduce the data demand for reliability monitoring of energy devices.
基金This work is supported by the National Natural Science Foundation of China(Nos.61771154,61603239,61772454,6171101570).
文摘Deep Learning(DL)is such a powerful tool that we have seen tremendous success in areas such as Computer Vision,Speech Recognition,and Natural Language Processing.Since Automated Modulation Classification(AMC)is an important part in Cognitive Radio Networks,we try to explore its potential in solving signal modulation recognition problem.It cannot be overlooked that DL model is a complex model,thus making them prone to over-fitting.DL model requires many training data to combat with over-fitting,but adding high quality labels to training data manually is not always cheap and accessible,especially in real-time system,which may counter unprecedented data in dataset.Semi-supervised Learning is a way to exploit unlabeled data effectively to reduce over-fitting in DL.In this paper,we extend Generative Adversarial Networks(GANs)to the semi-supervised learning will show it is a method can be used to create a more dataefficient classifier.
基金supported by the Fifth Key Project of Jiangsu Vocational Education Teaching Reform Research under Grant ZZZ13in part by the Science and Technology Project of Changzhou City under Grant CE20215032.
文摘Through semi-supervised learning and knowledge inheritance,a novel Takagi-Sugeno-Kang(TSK)fuzzy system framework is proposed for epilepsy data classification in this study.The new method is based on the maximum mean discrepancy(MMD)method and TSK fuzzy system,as a basic model for the classification of epilepsy data.First,formedical data,the interpretability of TSK fuzzy systems can ensure that the prediction results are traceable and safe.Second,in view of the deviation in the data distribution between the real source domain and the target domain,MMD is used to measure the distance between different data distributions.The objective function is constructed according to the MMD distance,and the distribution distance of different datasets is minimized to find the similar characteristics of different datasets.We introduce semi-supervised learning to further explore the relationship between data.Based on the MMD method,a semi-supervised learning(SSL)-MMD method is constructed by using pseudo-tags to realize the data distribution alignment of the same category.In addition,the idea of knowledge dissemination is used to learn pseudo-tags as additional data features.Finally,for epilepsy classification,the cross-domain TSK fuzzy system uses the cross-entropy function as the objective function and adopts the back-propagation strategy to optimize the parameters.The experimental results show that the new method can process complex epilepsy data and identify whether patients have epilepsy.
基金The publication of this article is funded by the Qatar National Library.
文摘Malaria is a lethal disease responsible for thousands of deaths worldwide every year.Manual methods of malaria diagnosis are timeconsuming that require a great deal of human expertise and efforts.Computerbased automated diagnosis of diseases is progressively becoming popular.Although deep learning models show high performance in the medical field,it demands a large volume of data for training which is hard to acquire for medical problems.Similarly,labeling of medical images can be done with the help of medical experts only.Several recent studies have utilized deep learning models to develop efficient malaria diagnostic system,which showed promising results.However,the most common problem with these models is that they need a large amount of data for training.This paper presents a computer-aided malaria diagnosis system that combines a semi-supervised generative adversarial network and transfer learning.The proposed model is trained in a semi-supervised manner and requires less training data than conventional deep learning models.Performance of the proposed model is evaluated on a publicly available dataset of blood smear images(with malariainfected and normal class)and achieved a classification accuracy of 96.6%.
基金supported by the Natural Science Foundation of China(No.51675089).
文摘In aerospace industry,gears are the most common parts of a mechanical transmission system.Gear pitting faults could cause the transmission system to crash and give rise to safety disaster.It is always a challenging problem to diagnose the gear pitting condition directly through the raw signal of vibration.In this paper,a novel method named augmented deep sparse autoencoder(ADSAE)is proposed.The method can be used to diagnose the gear pitting fault with relatively few raw vibration signal data.This method is mainly based on the theory of pitting fault diagnosis and creatively combines with both data augmentation ideology and the deep sparse autoencoder algorithm for the fault diagnosis of gear wear.The effectiveness of the proposed method is validated by experiments of six types of gear pitting conditions.The results show that the ADSAE method can effectively increase the network generalization ability and robustness with very high accuracy.This method can effectively diagnose different gear pitting conditions and show the obvious trend according to the severity of gear wear faults.The results obtained by the ADSAE method proposed in this paper are compared with those obtained by other common deep learning methods.This paper provides an important insight into the field of gear fault diagnosis based on deep learning and has a potential practical application value.
基金sponsored by the National Natural Science Foundation of China Grant No.62271302the Shanghai Municipal Natural Science Foundation Grant 20ZR1423500.
文摘Large amounts of labeled data are usually needed for training deep neural networks in medical image studies,particularly in medical image classification.However,in the field of semi-supervised medical image analysis,labeled data is very scarce due to patient privacy concerns.For researchers,obtaining high-quality labeled images is exceedingly challenging because it involves manual annotation and clinical understanding.In addition,skin datasets are highly suitable for medical image classification studies due to the inter-class relationships and the inter-class similarities of skin lesions.In this paper,we propose a model called Coalition Sample Relation Consistency(CSRC),a consistency-based method that leverages Canonical Correlation Analysis(CCA)to capture the intrinsic relationships between samples.Considering that traditional consistency-based models only focus on the consistency of prediction,we additionally explore the similarity between features by using CCA.We enforce feature relation consistency based on traditional models,encouraging the model to learn more meaningful information from unlabeled data.Finally,considering that cross-entropy loss is not as suitable as the supervised loss when studying with imbalanced datasets(i.e.,ISIC 2017 and ISIC 2018),we improve the supervised loss to achieve better classification accuracy.Our study shows that this model performs better than many semi-supervised methods.
文摘Machine learning techniques and a dataset of five wells from the Rawat oilfield in Sudan containing 93,925 samples per feature(seven well logs and one facies log) were used to classify four facies. Data preprocessing and preparation involve two processes: data cleaning and feature scaling. Several machine learning algorithms, including Linear Regression(LR), Decision Tree(DT), Support Vector Machine(SVM),Random Forest(RF), and Gradient Boosting(GB) for classification, were tested using different iterations and various combinations of features and parameters. The support vector radial kernel training model achieved an accuracy of 72.49% without grid search and 64.02% with grid search, while the blind-well test scores were 71.01% and 69.67%, respectively. The Decision Tree(DT) Hyperparameter Optimization model showed an accuracy of 64.15% for training and 67.45% for testing. In comparison, the Decision Tree coupled with grid search yielded better results, with a training score of 69.91% and a testing score of67.89%. The model's validation was carried out using the blind well validation approach, which achieved an accuracy of 69.81%. Three algorithms were used to generate the gradient-boosting model. During training, the Gradient Boosting classifier achieved an accuracy score of 71.57%, and during testing, it achieved 69.89%. The Grid Search model achieved a higher accuracy score of 72.14% during testing. The Extreme Gradient Boosting model had the lowest accuracy score, with only 66.13% for training and66.12% for testing. For validation, the Gradient Boosting(GB) classifier model achieved an accuracy score of 75.41% on the blind well test, while the Gradient Boosting with Grid Search achieved an accuracy score of 71.36%. The Enhanced Random Forest and Random Forest with Bagging algorithms were the most effective, with validation accuracies of 78.30% and 79.18%, respectively. However, the Random Forest and Random Forest with Grid Search models displayed significant variance between their training and testing scores, indicating the potential for overfitting. Random Forest(RF) and Gradient Boosting(GB) are highly effective for facies classification because they handle complex relationships and provide high predictive accuracy. The choice between the two depends on specific project requirements, including interpretability, computational resources, and data nature.
基金supported by China Southern Power Grid Co.Ltd.science and technology project(Research on the theory,technology and application of stereoscopic disaster defense for power distribution network in large city,GZHKJXM20180060)National Natural Science Foundation of China(No.51477100).
文摘It is crucial to maintain the safe and stable operation of distribution transformers,which constitute a key part of power systems.In the event of transformer failure,the fault type must be diagnosed in a timely and accurate manner.To this end,a transformer fault diagnosis method based on infrared image processing and semi-supervised learning is proposed herein.First,we perform feature extraction on the collected infrared-image data to extract temperature,texture,and shape features as the model reference vectors.Then,a generative adversarial network(GAN)is constructed to generate synthetic samples for the minority subset of labelled samples.The proposed method can learn information from unlabeled sample data,unlike conventional supervised learning methods.Subsequently,a semi-supervised graph model is trained on the entire dataset,i.e.,both labeled and unlabeled data.Finally,we test the proposed model on an actual dataset collected from a Chinese electricity provider.The experimental results show that the use of feature extraction,sample generation,and semi-supervised learning model can improve the accuracy of transformer fault classification.This verifies the effectiveness of the proposed method.
文摘The majority of big data analytics applied to transportation datasets suffer from being too domain-specific,that is,they draw conclusions for a dataset based on analytics on the same dataset.This makes models trained from one domain(e.g.taxi data)applies badly to a different domain(e.g.Uber data).To achieve accurate analyses on a new domain,substantial amounts of data must be available,which limits practical applications.To remedy this,we propose to use semi-supervised and active learning of big data to accomplish the domain adaptation task:Selectively choosing a small amount of datapoints from a new domain while achieving comparable performances to using all the datapoints.We choose the New York City(NYC)transportation data of taxi and Uber as our dataset,simulating different domains with 90%as the source data domain for training and the remaining 10%as the target data domain for evaluation.We propose semi-supervised and active learning strategies and apply it to the source domain for selecting datapoints.Experimental results show that our adaptation achieves a comparable performance of using all datapoints while using only a fraction of them,substantially reducing the amount of data required.Our approach has two major advantages:It can make accurate analytics and predictions when big datasets are not available,and even if big datasets are available,our approach chooses the most informative datapoints out of the dataset,making the process much more efficient without having to process huge amounts of data.
基金financially supported by the National Key R&D Program of China(No.2018YFA0702504)the National Natural Science Foundation of China(No.42174152 and No.41974140)+1 种基金the Science Foundation of China University of Petroleum,Beijing(No.2462020YXZZ008 and No.2462020QZDX003)the Strategic Cooperation Technology Projects of CNPC and CUPB(No.ZLZX2020-03).
文摘Intelligent seismic facies identification based on deep learning can alleviate the time-consuming and labor-intensive problem of manual interpretation,which has been widely applied.Supervised learning can realize facies identification with high efficiency and accuracy;however,it depends on the usage of a large amount of well-labeled data.To solve this issue,we propose herein an incremental semi-supervised method for intelligent facies identification.Our method considers the continuity of the lateral variation of strata and uses cosine similarity to quantify the similarity of the seismic data feature domain.The maximum-diff erence sample in the neighborhood of the currently used training data is then found to reasonably expand the training sets.This process continuously increases the amount of training data and learns its distribution.We integrate old knowledge while absorbing new ones to realize incremental semi-supervised learning and achieve the purpose of evolving the network models.In this work,accuracy and confusion matrix are employed to jointly control the predicted results of the model from both overall and partial aspects.The obtained values are then applied to a three-dimensional(3D)real dataset and used to quantitatively evaluate the results.Using unlabeled data,our proposed method acquires more accurate and stable testing results compared to conventional supervised learning algorithms that only use well-labeled data.A considerable improvement for small-sample categories is also observed.Using less than 1%of the training data,the proposed method can achieve an average accuracy of over 95%on the 3D dataset.In contrast,the conventional supervised learning algorithm achieved only approximately 85%.
基金Projects(61603393,61973306)supported in part by the National Natural Science Foundation of ChinaProject(BK20160275)supported by the Natural Science Foundation of Jiangsu Province,China+1 种基金Projects(2015M581885,2018T110571)supported by the Postdoctoral Science Foundation of ChinaProject(PAL-N201706)supported by the Open Project Foundation of State Key Laboratory of Synthetical Automation for Process Industries of Northeastern University,China
文摘Direct online measurement on product quality of industrial processes is difficult to be realized,which leads to a large number of unlabeled samples in modeling data.Therefore,it needs to employ semi-supervised learning(SSL)method to establish the soft sensor model of product quality.Considering the slow time-varying characteristic of industrial processes,the model parameters should be updated smoothly.According to this characteristic,this paper proposes an online adaptive semi-supervised learning algorithm based on random vector functional link network(RVFLN),denoted as OAS-RVFLN.By introducing a L2-fusion term that can be seen a weight deviation constraint,the proposed algorithm unifies the offline and online learning,and achieves smoothness of model parameter update.Empirical evaluations both on benchmark testing functions and datasets reveal that the proposed OAS-RVFLN can outperform the conventional methods in learning speed and accuracy.Finally,the OAS-RVFLN is applied to the coal dense medium separation process in coal industry to estimate the ash content of coal product,which further verifies its effectiveness and potential of industrial application.
基金supported by Korea Institute for Advancement of Technology(KIAT)grant funded by theKoreaGovernment(MOTIE)(P0008703,The CompetencyDevelopment Program for Industry Specialist).
文摘Intrusion detection involves identifying unauthorized network activity and recognizing whether the data constitute an abnormal network transmission.Recent research has focused on using semi-supervised learning mechanisms to identify abnormal network traffic to deal with labeled and unlabeled data in the industry.However,real-time training and classifying network traffic pose challenges,as they can lead to the degradation of the overall dataset and difficulties preventing attacks.Additionally,existing semi-supervised learning research might need to analyze the experimental results comprehensively.This paper proposes XA-GANomaly,a novel technique for explainable adaptive semi-supervised learning using GANomaly,an image anomalous detection model that dynamically trains small subsets to these issues.First,this research introduces a deep neural network(DNN)-based GANomaly for semi-supervised learning.Second,this paper presents the proposed adaptive algorithm for the DNN-based GANomaly,which is validated with four subsets of the adaptive dataset.Finally,this study demonstrates a monitoring system that incorporates three explainable techniques—Shapley additive explanations,reconstruction error visualization,and t-distributed stochastic neighbor embedding—to respond effectively to attacks on traffic data at each feature engineering stage,semi-supervised learning,and adaptive learning.Compared to other single-class classification techniques,the proposed DNN-based GANomaly achieves higher scores for Network Security Laboratory-Knowledge Discovery in Databases and UNSW-NB15 datasets at 13%and 8%of F1 scores and 4.17%and 11.51%for accuracy,respectively.Furthermore,experiments of the proposed adaptive learning reveal mostly improved results over the initial values.An analysis and monitoring system based on the combination of the three explainable methodologies is also described.Thus,the proposed method has the potential advantages to be applied in practical industry,and future research will explore handling unbalanced real-time datasets in various scenarios.
基金supported in part by the National Key R&D Program of China under Grant 2018YFA0701601part by the National Natural Science Foundation of China(Grant No.U22A2002,61941104,62201605)part by Tsinghua University-China Mobile Communications Group Co.,Ltd.Joint Institute。
文摘In the upcoming large-scale Internet of Things(Io T),it is increasingly challenging to defend against malicious traffic,due to the heterogeneity of Io T devices and the diversity of Io T communication protocols.In this paper,we propose a semi-supervised learning-based approach to detect malicious traffic at the access side.It overcomes the resource-bottleneck problem of traditional malicious traffic defenders which are deployed at the victim side,and also is free of labeled traffic data in model training.Specifically,we design a coarse-grained behavior model of Io T devices by self-supervised learning with unlabeled traffic data.Then,we fine-tune this model to improve its accuracy in malicious traffic detection by adopting a transfer learning method using a small amount of labeled data.Experimental results show that our method can achieve the accuracy of 99.52%and the F1-score of 99.52%with only 1%of the labeled training data based on the CICDDoS2019 dataset.Moreover,our method outperforms the stateof-the-art supervised learning-based methods in terms of accuracy,precision,recall and F1-score with 1%of the training data.
基金supported by Innovation Talents Promotion Program of Shaanxi Province,China(No.2021TD08)。
文摘Radio frequency fingerprinting(RFF)is a remarkable lightweight authentication scheme to support rapid and scalable identification in the internet of things(IoT)systems.Deep learning(DL)is a critical enabler of RFF identification by leveraging the hardware-level features.However,traditional supervised learning methods require huge labeled training samples.Therefore,how to establish a highperformance supervised learning model with few labels under practical application is still challenging.To address this issue,we in this paper propose a novel RFF semi-supervised learning(RFFSSL)model which can obtain a better performance with few meta labels.Specifically,the proposed RFFSSL model is constituted by a teacher-student network,in which the student network learns from the pseudo label predicted by the teacher.Then,the output of the student model will be exploited to improve the performance of teacher among the labeled data.Furthermore,a comprehensive evaluation on the accuracy is conducted.We derive about 50 GB real long-term evolution(LTE)mobile phone’s raw signal datasets,which is used to evaluate various models.Experimental results demonstrate that the proposed RFFSSL scheme can achieve up to 97%experimental testing accuracy over a noisy environment only with 10%labeled samples when training samples equal to 2700.
基金the National Key Technology Research and Development Program of China(No.2015BAH13F01)the Beijing Natural Science Foundation(No.4152007)
文摘The rapid development of the Internet brings a variety of original information including text information, audio information, etc. However, it is difficult to find the most useful knowledge rapidly and accurately because of its huge number. Automatic text classification technology based on machine learning can classify a large number of natural language documents into the corresponding subject categories according to its correct semantics. It is helpful to grasp the text information directly. By learning from a set of hand-labeled documents,we obtain the traditional supervised classifier for text categorization(TC). However, labeling all data by human is labor intensive and time consuming. To solve this problem, some scholars proposed a semi-supervised learning method to train classifier, but it is unfeasible for various kinds and great number of Web data since it still needs a part of hand-labeled data. In 2012, Li et al. invented a fully automatic categorization approach for text(FACT)based on supervised learning, where no manual labeling efforts are required. But automatically labeling all data can bring noise into experiment and cause the fact that the result cannot meet the accuracy requirement. We put forward a new idea that part of data with high accuracy can be automatically tagged based on the semantic of category name, then a semi-supervised way is taken to train classifier with both labeled and unlabeled data,and ultimately a precise classification of massive text data can be achieved. The empirical experiments show that the method outperforms the supervised support vector machine(SVM) in terms of both F1 performance and classification accuracy in most cases. It proves the effectiveness of the semi-supervised algorithm in automatic TC.
基金This research was supported by the BK21 FOUR(Fostering Outstanding Universities for Research)funded by the Ministry of Education(MOE,Korea)and National Research Foundation of Korea(NFR).
文摘Artificial intelligent based dialog systems are getting attention from both business and academic communities.The key parts for such intelligent chatbot systems are domain classification,intent detection,and named entity recognition.Various supervised,unsupervised,and hybrid approaches are used to detect each field.Such intelligent systems,also called natural language understanding systems analyze user requests in sequential order:domain classification,intent,and entity recognition based on the semantic rules of the classified domain.This sequential approach propagates the downstream error;i.e.,if the domain classification model fails to classify the domain,intent and entity recognition fail.Furthermore,training such intelligent system necessitates a large number of user-annotated datasets for each domain.This study proposes a single joint predictive deep neural network framework based on long short-term memory using only a small user-annotated dataset to address these issues.It investigates value added by incorporating unlabeled data from user chatting logs into multi-domain spoken language understanding systems.Systematic experimental analysis of the proposed joint frameworks,along with the semi-supervised multi-domain model,using open-source annotated and unannotated utterances shows robust improvement in the predictive performance of the proposed multi-domain intelligent chatbot over a base joint model and joint model based on adversarial learning.
基金This work was supported by the National Research Foundation of Korea(No.2020R1A2C1014829)by the Korea Medical Device Development Fund grant,which is funded by the Government of the Republic of Korea Korea government(the Ministry of Science and ICT+2 种基金the Ministry of Trade,Industry and Energythe Ministry of Health and Welfareand the Ministry of Food and Drug Safety)(grant KMDF_PR_20200901_0095).
文摘For the classification problem in practice,one of the challenging issues is to obtain enough labeled data for training.Moreover,even if such labeled data has been sufficiently accumulated,most datasets often exhibit long-tailed distribution with heavy class imbalance,which results in a biased model towards a majority class.To alleviate such class imbalance,semisupervised learning methods using additional unlabeled data have been considered.However,as a matter of course,the accuracy is much lower than that from supervised learning.In this study,under the assumption that additional unlabeled data is available,we propose the iterative semi-supervised learning algorithms,which iteratively correct the labeling of the extra unlabeled data based on softmax probabilities.The results show that the proposed algorithms provide the accuracy as high as that from the supervised learning.To validate the proposed algorithms,we tested on the two scenarios:with the balanced unlabeled dataset and with the imbalanced unlabeled dataset.Under both scenarios,our proposed semi-supervised learning algorithms provided higher accuracy than previous state-of-the-arts.Code is available at https://github.com/HeewonChung92/iterative-semi-learning.
文摘Significant breakthroughs in medical image registration have been achieved using deep neural networks(DNNs).However,DNN-based end-to-end registration methods often require large quantities of data or adequate annotations for training.To leverage the intensity information of abundant unlabeled images,unsupervised registration methods commonly employ intensity-based similarity measures to optimize the network parameters.However,finding a sufficiently robust measure can be challenging for specific registration applications.Weakly supervised registration methods use anatomical labels to estimate the deformation between images.High-level structural information in label images is more reliable and practical for estimating the voxel correspondence of anatomic regions of interest between images,whereas label images are extremely difficult to collect.In this paper,we propose a two-stage semi-supervised learning framework for medical image registration,which consists of unsupervised and weakly supervised registration networks.The proposed semi-supervised learning framework is trained with intensity information from available images,label information from a relatively small number of labeled images and pseudo-label information from unlabeled images.Experimental results on two datasets(cardiac and abdominal images)demonstrate the efficacy and efficiency of this method in intra-and inter-modality medical image registrations,as well as its superior performance when a vast amount of unlabeled data and a small set of annotations are available.Our code is publicly available at at https://github.com/jdq818/SeRN.
