Unlabeled Data In Target Domain Research Articles

Rectifying Pseudo Label Learning via Uncertainty Estimation for Domain Adaptive Semantic Segmentation

This paper focuses on the unsupervised domain adaptation of transferring the knowledge from the source domain to the target domain in the context of semantic segmentation. Existing approaches usually regard the pseudo label as the ground truth to fully exploit the unlabeled target-domain data. Yet the pseudo labels of the target-domain data are usually predicted by the model trained on the source domain. Thus, the generated labels inevitably contain the incorrect prediction due to the discrepancy between the training domain and the test domain, which could be transferred to the final adapted model and largely compromises the training process. To overcome the problem, this paper proposes to explicitly estimate the prediction uncertainty during training to rectify the pseudo label learning for unsupervised semantic segmentation adaptation. Given the input image, the model outputs the semantic segmentation prediction as well as the uncertainty of the prediction. Specifically, we model the uncertainty via the prediction variance and involve the uncertainty into the optimization objective. To verify the effectiveness of the proposed method, we evaluate the proposed method on two prevalent synthetic-to-real semantic segmentation benchmarks, i.e., GTA5 \(\rightarrow \) Cityscapes and SYNTHIA \(\rightarrow \) Cityscapes, as well as one cross-city benchmark, i.e., Cityscapes \(\rightarrow \) Oxford RobotCar. We demonstrate through extensive experiments that the proposed approach (1) dynamically sets different confidence thresholds according to the prediction variance, (2) rectifies the learning from noisy pseudo labels, and (3) achieves significant improvements over the conventional pseudo label learning and yields competitive performance on all three benchmarks.

IFDS: An Intelligent Fault Diagnosis System With Multisource Unsupervised Domain Adaptation for Different Working Conditions

Intelligent fault diagnosis is an important subject of mechanical system maintenance. Domain adaptation is a method to solve the problem that the model trained on the training set (source domain) is not suitable for the test set (target domain) due to different working conditions in fault diagnosis. In industrial scenarios, there may be multiple-source domains. For this reason, we proposed an intelligent fault diagnosis system (IFDS) with multisource unsupervised domain adaptive network that adapts to single- or multiple-source domains. The proposed method considers the differences between sources and uses source domain data and a small amount of unlabeled target domain data to mine the feature information contained in the data. IFDS uses a feature extractor to learn the feature representations, constructs a domain discriminator for each source domain, and learns domain-invariant features through adversarial training to diagnose target domain faults. The validity of the method in fault diagnosis is verified by various transfer tasks of two public fault-bearing datasets.

Learning Causal Representations for Robust Domain Adaptation

Domain adaptation solves the learning problem in a target domain by leveraging the knowledge in a relevant source domain. While remarkable advances have been made, almost all existing domain adaptation methods heavily require large amounts of unlabeled target domain data for learning domain invariant representations to achieve good generalizability on the target domain. In fact, in many real-world applications, target domain data may not always be available. In this paper, we study the cases where at the training phase the target domain data is unavailable and only well-labeled source domain data is available, called robust domain adaptation. To tackle this problem, under the assumption that causal relationships between features and the class variable are robust across domains, we propose a novel Causal AutoEncoder (CAE), which integrates deep autoencoder and causal structure learning into a unified model to learn causal representations only using data from a single source domain. Specifically, a deep autoencoder model is adopted to learn low-dimensional representations, and a causal structure learning model is designed to separate the low-dimensional representations into two groups: causal representations and task-irrelevant representations. Using three real-world datasets the extensive experiments have validated the effectiveness of CAE compared to eleven state-of-the-art methods.

Graph Domain Adversarial Transfer Network for Cross-Domain Sentiment Classification

In the text sentiment classification task, some words are seemingly unrelated to the classification task, but they have a direct impact on the performance of classification model. For example, in the sentences “I have terminal cancer” and “Cancer is a very common disease”, it can be clearly found that the word “cancer” has two different sentiment tendencies in the daily life domain and the medical domain. In the daily life domain, the word “cancer” shows an extremely negative sentiment tendency. While in the medical domain, the word “cancer” is just a simple term with a relatively neutral sentiment tendency. Although current deep learning models have already achieved good performance through their powerful feature learning capabilities, there are serious deficiencies in dealing with the above problem. Therefore, from a new perspective, this paper proposes the Graph Domain Adversarial Transfer Network (GDATN) based on the idea of adversarial learning, which uses the labeled source domain data to predict the sentiment label of unlabeled target domain data. Firstly, GDATN extracts feature representations through the Bidirectional Long Short-Term Memory (BiLSTM) Network and Graph Attention Network (GAT) successively. Then, GDATN introduces the domain classifier to capture the domain-shared text feature representation with the Gradient Reversal Layer (GRL). In addition, an auxiliary task named the projection mechanism is constructed to further capture the domain-specific text feature representation in response to the text domain problem. Extensive experimental results on two benchmark datasets show that GDATN proposed in this paper outperforms the other six benchmark sentiment classification models, and GDATN has a better stability on different cross-domain pairs.

Minimizing-Entropy and Fourier Consistency Network for Domain Adaptation on Optic Disc and Cup Segmentation

Automated segmentation of the optic disc (OD) and optic cup (OC) from different datasets plays an important role in the diagnosis of glaucoma and greatly saves human resources in both data annotation and image segmentation. However, the domain shift between different datasets suppresses the generalization ability of the segmentation network, especially damaging the performance of segmentation in the target domain, which is unlabeled. Therefore, using a transfer learning algorithm or domain adaptation method to enhance the migration ability of segmentation models has become an essential step and has attracted the attention of many researchers. In this paper, we propose an unsupervised domain adaptation network, called the Minimizing-entropy and Fourier Domain Adaptation network (MeFDA), to narrow the discrepancy between the source and target domains and prevent the degradation of segmentation performance. First, we perform adversarial optimization on the entropy maps of the predicted segmentation results to alleviate the domain shift. Then, direct entropy-minimization optimization is applied to the unlabeled target domain data to improve the credibility of the prediction segmentation maps. To enhance the prediction consistency of the target domain data, we augment the target domain dataset through the Fourier transform by replacing the low-frequency part in the target images with that of the source images. Then, a semantic consistency constraint is imposed on the raw images and augmented images of the target domain to improve the prediction consistency of the segmentation model, thereby further narrowing the discrepancy between the source and target domains. Experiments on several public retinal fundus image datasets prove the superiority of MeFDA compared with state-of-the-art methods, and the ablation study analyzes the importance of the different proposed components.

A new transferable bearing fault diagnosis approach with adaptive manifold embedded distribution alignment

The conditional and marginal distributions of bearing vibration signal data collected under different working conditions generally may not obey the same distribution. Moreover, feature distortions are difficult to eliminate when performing distribution alignment in the original space. Based on these challenges, this study proposes a novel transferable fault diagnosis approach with adaptive manifold embedded distribution alignment (AMEDA). AMEDA can learn transformed feature representations in the Grassmann manifold space by constructing a geodesic flow kernel to avoid feature distortions. In addition, this paper initially aims to construct an adaptive factor defined by -distance to adjust the relative importance of the conditional and marginal distributions dynamically. After manifold feature learning and dynamic distribution alignment, a cross-domain classifier f with structural risk minimization can be learned to predict unlabeled target domain data. Experimental results from multiple indicators based on two datasets demonstrate the superiority of AMEDA over the other existing methods.

Compact class-conditional domain invariant learning for multi-class domain adaptation

Neural network-based models have recently shown excellent performance in various kinds of tasks. However, a large amount of labeled data is required to train deep networks, and the cost of gathering labeled training data for every kind of domain is prohibitively expensive. Domain adaptation tries to solve this problem by transferring knowledge from labeled source domain data to unlabeled target domain data. Previous research tried to learn domain-invariant features of source and target domains to address this problem, and this approach has been used as a key concept in various methods. However, domain-invariant features do not mean that a classifier trained on source data can be directly applied to target data because it does not guarantee that data distribution of the same classes will be aligned across two domains. In this paper, we present novel generalization upper bounds for domain adaptation that motivates the need for class-conditional domain invariant learning. Based on this theoretical framework, we then propose a class-conditional domain invariant learning method that can learn a feature space in which features in the same class are expected to be mapped nearby. We empirically experimented that our model showed state-of-the-art performance on standard datasets and showed effectiveness by visualization of latent space.

TScatNet: An Interpretable Cross-Domain Intelligent Diagnosis Model With Antinoise and Few-Shot Learning Capability

In a real industrial scenario, domain shift frequently occurred due to working loads variation, operation speeds variation, and environmental noise interference, severely degrading intelligent fault diagnosis models' performance. Currently, domain adaptation-based models eliminate domain shift by calibrating unlabeled target-domain data using labeled source-domain data. Nevertheless, these models may fail when encountering unseen working conditions, lacking unlabeled target-domain data for learning domain-invariant features. Besides, the existing deep domain adaptation-based models lack a few-shot learning capability and interpretability. This article develops a cross-domain diagnosis model named time-scattering convolutional network (TScatNet) to remedy these gaps. TScatNet extracts domain-invariant features using Morlet wavelet as the predefined convolutional kernel, modulus as nonlinearity, and scaling averaging as pooling layer. This predefined architecture eliminates domain shift without any domain adaptation, endows TScatNet few-shot learning capability, simplifies the hyperparameter tuning process, and brings interpretability. Both the CWRU and DDS data sets are used to verify the proposed model, which shows that TScatNet could stably realize 100% accuracy on transfer tasks across working loads and 100% accuracy across operation speeds. Moreover, even though the SNR value descends to -4, TScatNet achieved 96% accuracy on ten categories tasks and 99.8% accuracy on four categories tasks. Besides, TScatNet achieved nearly 100% accuracy both under training samples' sparsity and sparsity of working conditions.

Domain-adaptive intelligence for fault diagnosis based on deep transfer learning from scientific test rigs to industrial applications

With the accumulation of data, the intelligent fault diagnosis of rolling bearings has achieved fruitful results, but it is costly to acquire and label data for industrial application. A series of studies achieve the reuse of knowledge from bearings in scientific test rigs (BSTRs) to bearings in industrial applications (BIAs) via transfer learning. Nevertheless, most of the cases ignore a constraint: the industrial datasets suffers from class imbalance. In general, industrial datasets lack data samples of fault states. To this end, we propose a pseudo-categorized maximum mean discrepancy (PCMMD), and use it to drive the multi-input multi-output convolutional network (MIMOCN) to narrow the cross-domain distribution discrepancy of various categories in the deep feature space. Firstly, the domain-shared encoder and classifier are pre-trained based on the source-domain labeled dataset. Then, the labeled source-domain data and unlabeled target-domain data are jointly used to train the MIMOCN. The pseudo-labels enable the PCMMD to measure the intra-class cross-domain distribution discrepancy. The target-domain reconstruction object function improves the credibility of target-domain deep feature. The transfer cases from artificial damaged BSTRs to real damage BSTRs verify the stability of the proposed PCMMD. In the case of transfer learning from balanced BSTRs to BIAs lacking fault samples, the diagnosis accuracy of BIAs is higher than that of existing methods.

Few shot domain adaptation for in situ macromolecule structural classification in cryoelectron tomograms.

Cryoelectron tomography (cryo-ET) visualizes structure and spatial organization of macromolecules and their interactions with other subcellular components inside single cells in the close-to-native state at submolecular resolution. Such information is critical for the accurate understanding of cellular processes. However, subtomogram classification remains one of the major challenges for the systematic recognition and recovery of the macromolecule structures in cryo-ET because of imaging limits and data quantity. Recently, deep learning has significantly improved the throughput and accuracy of large-scale subtomogram classification. However, often it is difficult to get enough high-quality annotated subtomogram data for supervised training due to the enormous expense of labeling. To tackle this problem, it is beneficial to utilize another already annotated dataset to assist the training process. However, due to the discrepancy of image intensity distribution between source domain and target domain, the model trained on subtomograms in source domain may perform poorly in predicting subtomogram classes in the target domain. In this article, we adapt a few shot domain adaptation method for deep learning-based cross-domain subtomogram classification. The essential idea of our method consists of two parts: (i) take full advantage of the distribution of plentiful unlabeled target domain data, and (ii) exploit the correlation between the whole source domain dataset and few labeled target domain data. Experiments conducted on simulated and real datasets show that our method achieves significant improvement on cross domain subtomogram classification compared with baseline methods. Software is available online https://github.com/xulabs/aitom. Supplementary data are available at Bioinformatics online.

A New Deep Convolutional Domain Adaptation Network for Bearing Fault Diagnosis under Different Working Conditions

Effective fault diagnosis methods can ensure the safe and reliable operation of the machines. In recent years, deep learning technology has been applied to diagnose various mechanical equipment faults. However, in real industries, the data distribution under different working conditions is often different, which leads to serious degradation of diagnostic performance. In order to solve the issue, this study proposes a new deep convolutional domain adaptation network (DCDAN) method for bearing fault diagnosis. This method implements cross-domain fault diagnosis by using the labeled source domain data and the unlabeled target domain data as training data. In DCDAN, firstly, a convolutional neural network is applied to extract features of source domain data and target domain data. Then, the domain distribution discrepancy is reduced through minimizing probability distribution distance of multiple kernel maximum mean discrepancies (MK-MMD) and maximizing the domain recognition error of domain classifier. Finally, the source domain classification error is minimized. Extensive experiments on two rolling bearing datasets verify that the proposed method can implement accurate cross-domain fault diagnosis under different working conditions. The study may provide a promising tool for bearing fault diagnosis under different working conditions.

Wavefront reconstruction based on deep transfer learning for microscopy.

The application of machine learning in wavefront reconstruction has brought great benefits to real-time, non-invasive, deep tissue imaging in biomedical research. However, due to the diversity and heterogeneity of biological tissues, it is difficult to train the dataset with a unified model. In general, the utilization of some unified models will result in the specific sample falling outside the training set, leading to low accuracy of the machine learning model in some real applications. This paper proposes a sensorless wavefront reconstruction method based on transfer learning to overcome the domain shift introduced by the difference between the training set and the target test set. We build a weights-sharing two-stream convolutional neural network (CNN) framework for the prediction of Zernike coefficient, in which a large number of labeled randomly generated samples serve as the source-domain data and the unlabeled specific samples serve as the target-domain data at the same time. By training on massive labeled simulated data with domain adaptation to unlabeled target-domain data, the network shows better performance on the target tissue samples. Experimental results show that the accuracy of the proposed method is 18.5% higher than that of conventional CNN-based method and the peak intensities of the point spread function (PSF) are more than 20% higher with almost the same training time and processing time. The better compensation performance on target sample could have more advantages when handling complex aberrations, especially the aberrations caused by various histological characteristics, such as refractive index inhomogeneity and biological motion in biological tissues.

Enhancing unsupervised domain adaptation by discriminative relevance regularization

Unsupervised domain adaptation (UDA) serves to transfer specific knowledge from massive labeled source domain data to unlabeled target domain data via mitigating domain shift. In this paper, we propose a discriminative relevance regularization term (DRR) to enhance the performance of UDA by reducing the domain shift from the aspect of semantic relevance across domains. In particular, DRR is formulated as the min–max rank problem which seeks a projection matrix to minimize the rank of intra-class projected features and maximize the rank of the means of inter-class projected features simultaneously. To test the potential effectiveness of DRR, we design a relevance regularized distribution adaptation method (RRDA) and relevance regularized adaptation networks (RRAN) for image classification, and a relevance regularized self-supervised learning method (RRSL) for semantic segmentation by incorporation of DRR. The corresponding optimization algorithms are proposed to solve them. Experiments of cross-domain image classification show that both RRDA and RRAN outperform several state-of-the-art compared methods. Moreover, experiments of domain-adaptation semantic segmentation on two synthetic-to-real segmentation datasets demonstrate the capacity of RRSL. Such results imply the efficacy of DRR on both image classification and semantic segmentation tasks.

Selective transfer cycle GAN for unsupervised person re-identification

Recently, many researchers pay more attention to unsupervised person re-identification (UPRID) due to its scalability and flexibility without any labeled data. The popular methods for UPRID are divided into clustering based and transfer learning based approaches. However, both of them either lack pre-knowledge or some useless knowledge transfered, which always achieves poor matching performance. Specifically, transfer learning based methods can utilize existing labeled data to boost the training of unlabeled data in target domain. Nevertheless, they generally offer weaker re-identification performance because there are a lot of negative images in source domain bringing negative effect on accuracy for the target domain, which is very challenging to address this negative source data removing problem. In this paper, we propose a Selective Transfer Cycle Generative Adversarial Network (STCGAN) by selecting “valuable” source domain knowledge to boost the training efficiency in unlabeled target domain. Our STCGAN approach is developed from a Cycle GAN attached a selector for source data, a part based feature extractor for target data and reconstructed source images following target distribution. Such that it can simultaneously learn “valuable” source images through the selector and exploit transferable discriminative information from these selected source images into target domain. Then, we introduce a joint optimization method and conduct extensively experiments on two widely used person re-identification datasets. The results show the superiority of the proposed STCGAN model over a range of the-state-of-the-arts.

Transfer Neural Trees: Semi-Supervised Heterogeneous Domain Adaptation and Beyond.

Heterogeneous domain adaptation (HDA) addresses the task of associating data not only across dissimilar domains but also described by different types of features. Inspired by the recent advances of neural networks and deep learning, we propose a deep leaning model of Transfer Neural Trees (TNT), which jointly solves cross-domain feature mapping, adaptation, and classification in a unified architecture. As the prediction layer in TNT, we introduce Transfer Neural Decision Forest (Transfer- NDF), which is able to learn the neurons in TNT for adaptation by stochastic pruning. In order to handle semi-supervised HDA, a unique embedding loss term is introduced to TNT for preserving prediction and structural consistency between labeled and unlabeled target-domain data. We further show that our TNT can be extended to zero shot learning for associating image and attribute data with promising performance. Finally, experiments on different classification tasks across features, datasets, and modalities would verify the effectiveness of our TNT.

Discriminative transfer learning via local and global structure preservation

The current success of supervised learning is limited on large amounts of labeled training data. Transfer learning aims to learn an adaptive classifier for the unlabeled target domain data from the labeled source domain data, which is sampled from diverse probability distributions under changing conditions. Most previous works focus on how to reduce the distribution discrepancy between two involved domains, or exploit the shared common feature by preserving the local geometric structure of samples. In this paper, we propose a modified method jointly optimizing the local and global structure preservation. The main idea is to explore common features with manifold regularization. Discriminative repulsive force model is used to improve maximum mean discrepancy, which keeps discriminative property in the local sense via labeled source domain data and alleviates the global distribution discrepancy of the different domains. Quantitative results indicate that our method performs better than other methods on 16 cross-domain experiments.

Semi-supervised domain adaptation via Fredholm integral based kernel methods

Along with the emergence of domain adaptation in semi-supervised setting, dealing with the noisy and complex data in classifier adaptation underscores its growing importance. We believe a large amount of unlabeled data in target domain, which are always only used in distribution alignment, are more of a great source of information for this challenge. In this paper, we propose a novel Transfer Fredholm Multiple Kernel Learning (TFMKL) framework to suppress the noise for complex data distributions. Firstly, with exploring unlabeled target data, TFMKL learns a cross-domain predictive model by developing a Fredholm integral based kernel prediction framework which is proven to be effective in noise suppression. Secondly, TFMKL explicitly extends the applied range of unlabeled target samples into adaptive classifier building and distribution alignment. Thirdly, multiple kernels are explored to induce an optimal learning space. Correspondingly, TFMKL is distinguished with allowing for noise resiliency, facilitating knowledge transfer and analyzing complex data characteristics at the same time. Furthermore, an effective optimization procedure is presented based on the reduced gradient, guaranteeing rapid convergence. We emphasize the adaptability of TFMKL to different domain adaptation tasks due to its extension to different predictive models. In particular, two models based on square loss and hinge loss respectively are proposed within the TFMKL framework. Comprehensive empirical studies on benchmark data sets verify the effectiveness and the noise resiliency of our proposed methods.

Epithelium-Stroma Classification via Convolutional Neural Networks and Unsupervised Domain Adaptation in Histopathological Images.

Epithelium-stroma classification is a necessary preprocessing step in histopathological image analysis. Current deep learning based recognition methods for histology data require collection of large volumes of labeled data in order to train a new neural network when there are changes to the image acquisition procedure. However, it is extremely expensive for pathologists to manually label sufficient volumes of data for each pathology study in a professional manner, which results in limitations in real-world applications. A very simple but effective deep learning method, that introduces the concept of unsupervised domain adaptation to a simple convolutional neural network (CNN), has been proposed in this paper. Inspired by transfer learning, our paper assumes that the training data and testing data follow different distributions, and there is an adaptation operation to more accurately estimate the kernels in CNN in feature extraction, in order to enhance performance by transferring knowledge from labeled data in source domain to unlabeled data in target domain. The model has been evaluated using three independent public epithelium-stroma datasets by cross-dataset validations. The experimental results demonstrate that for epithelium-stroma classification, the proposed framework outperforms the state-of-the-art deep neural network model, and it also achieves better performance than other existing deep domain adaptation methods. The proposed model can be considered to be a better option for real-world applications in histopathological image analysis, since there is no longer a requirement for large-scale labeled data in each specified domain.

An Exemplar-Based Multi-View Domain Generalization Framework for Visual Recognition.

In this paper, we propose a new exemplar-based multi-view domain generalization (EMVDG) framework for visual recognition by learning robust classifier that are able to generalize well to arbitrary target domain based on the training samples with multiple types of features (i.e., multi-view features). In this framework, we aim to address two issues simultaneously. First, the distribution of training samples (i.e., the source domain) is often considerably different from that of testing samples (i.e., the target domain), so the performance of the classifiers learnt on the source domain may drop significantly on the target domain. Moreover, the testing data are often unseen during the training procedure. Second, when the training data are associated with multi-view features, the recognition performance can be further improved by exploiting the relation among multiple types of features. To address the first issue, considering that it has been shown that fusing multiple SVM classifiers can enhance the domain generalization ability, we build our EMVDG framework upon exemplar SVMs (ESVMs), in which a set of ESVM classifiers are learnt with each one trained based on one positive training sample and all the negative training samples. When the source domain contains multiple latent domains, the learnt ESVM classifiers are expected to be grouped into multiple clusters. To address the second issue, we propose two approaches under the EMVDG framework based on the consensus principle and the complementary principle, respectively. Specifically, we propose an EMVDG_CO method by adding a co-regularizer to enforce the cluster structures of ESVM classifiers on different views to be consistent based on the consensus principle. Inspired by multiple kernel learning, we also propose another EMVDG_MK method by fusing the ESVM classifiers from different views based on the complementary principle. In addition, we further extend our EMVDG framework to exemplar-based multi-view domain adaptation (EMVDA) framework when the unlabeled target domain data are available during the training procedure. The effectiveness of our EMVDG and EMVDA frameworks for visual recognition is clearly demonstrated by comprehensive experiments on three benchmark data sets.

Unsupervised Domain Adaptation With Label and Structural Consistency.

Unsupervised domain adaptation deals with scenarios in which labeled data are available in the source domain, but only unlabeled data can be observed in the target domain. Since the classifiers trained by source-domain data would not be expected to generalize well in the target domain, how to transfer the label information from source to target-domain data is a challenging task. A common technique for unsupervised domain adaptation is to match cross-domain data distributions, so that the domain and distribution differences can be suppressed. In this paper, we propose to utilize the label information inferred from the source domain, while the structural information of the unlabeled target-domain data will be jointly exploited for adaptation purposes. Our proposed model not only reduces the distribution mismatch between domains, improved recognition of target-domain data can be achieved simultaneously. In the experiments, we will show that our approach performs favorably against the state-of-the-art unsupervised domain adaptation methods on benchmark data sets. We will also provide convergence, sensitivity, and robustness analysis, which support the use of our model for cross-domain classification.