Cross-domain Classification Research Articles

Cycle label-consistent networks for unsupervised domain adaptation

Domain adaptation aims to leverage a labeled source domain to learn a classifier for the unlabeled target domain with a different distribution. Previous methods mostly match the distribution between two domains by global or class alignment. However, global alignment methods cannot achieve a fine-grained class-to-class overlap; class alignment methods supervised by pseudo-labels cannot guarantee their reliability. In this paper, we propose a simple yet efficient domain adaptation method, i.e. Cycle Label-Consistent Network (CLCN), by exploiting the cycle consistency of classification label, which applies dual cross-domain nearest centroid classification procedures to generate a reliable self-supervised signal for the discrimination in the target domain. The cycle label-consistent loss reinforces the consistency between ground-truth labels and pseudo-labels of source samples leading to statistically similar latent representations between source and target domains. This new loss can easily be added to any existing classification network with almost no computational overhead. We demonstrate the effectiveness of our approach on MNIST-USPS-SVHN, Office-31, Office-Home and Image CLEF-DA benchmarks. Results validate that the proposed method can alleviate the negative influence of falsely-labeled samples and learn more discriminative features, leading to the absolute improvement over source-only model by 9.4% on Office-31 and 6.3% on Image CLEF-DA.

Misogyny Detection in Twitter: a Multilingual and Cross-Domain Study

The freedom of expression given by social media has a dark side: the growing proliferation of abusive contents on these platforms. Misogynistic speech is a kind of abusive language, which can be simplified as hate speech targeting women, and it is becoming a more and more relevant issue in recent years. AMI IberEval 2018 and AMI EVALITA 2018 were two shared tasks which mainly focused on tackling the problem of misogyny in Twitter, in three different languages, namely English, Italian, and Spanish. In this paper, we present an in-depth study on the phenomena of misogyny in those three languages, by focusing on three main objectives. Firstly, we investigate the most important features to detect misogyny and the issues which contribute to the difficulty of misogyny detection, by proposing a novel system and conducting a broad evaluation on this task. Secondly, we study the relationship between misogyny and other abusive language phenomena, by conducting a series of cross-domain classification experiments. Finally, we explore the feasibility of detecting misogyny in a multilingual environment, by carrying out cross-lingual classification experiments. Our system succeeded to outperform all state of the art systems in all benchmark AMI datasets both subtask A and subtask B. Moreover, intriguing insights emerged from error analysis, in particular about the interaction between different but related abusive phenomena. Based on our cross-domain experiment, we conclude that misogyny is quite a specific kind of abusive language, while we experimentally found that it is different from sexism. Lastly, our cross-lingual experiments show promising results. Our proposed joint-learning architecture obtained a robust performance across languages, worth to be explored in further investigation.

Unified Cross-domain Classification via Geometric and Statistical Adaptations

Domain adaptation aims to learn an adaptive classifier for target data using the labelled source data from a different distribution. Most proposed works construct cross-domain classifier by exploring one-sided property of the input data, i.e., either geometric or statistical property. Therefore they may ignore the complementarity between the two properties. Moreover, many previous methods implement knowledge transfer with two separated steps: divergence minimization and classifier construction, which degrades the adaptation robustness. In order to address such problems, we propose a unified cross-domain classification method via geometric and statistical adaptations (UCGS). UCGS models the divergence minimization and classifier construction in a unified way based on structural risk minimization principle and coupled adaptations theory. Specifically, UCGS constructs an adaptive model by simultaneously minimizing the structural risk on labelled source data, using Maximum Mean Discrepancy (MMD) criterion to implement statistical adaptation, and flexibly employing the Nyström method to explore the geometric connections between domains. A domain-invariant graph is successfully constructed to link the two domains geometrically. The standard supervised methods can be used to instantiate UCGS to handle inter-domain classification problems. Comprehensive experiments show the superiority of UCGS on several real-world datasets.

Effective Visual Domain Adaptation via Generative Adversarial Distribution Matching.

In the field of computer vision, without sufficient labeled images, it is challenging to train an accurate model. However, through visual adaptation from source to target domains, a relevant labeled dataset can help solve such problem. Many methods apply adversarial learning to diminish cross-domain distribution difference. They are able to greatly enhance the performance on target classification tasks. Generative adversarial network (GAN) loss is widely used in adversarial adaptation learning methods to reduce an across-domain distribution difference. However, it becomes difficult to decline such distribution difference if generator or discriminator in GAN fails to work as expected and degrades its performance. To solve such cross-domain classification problems, we put forward a novel adaptation framework called generative adversarial distribution matching (GADM). In GADM, we improve the objective function by taking cross-domain discrepancy distance into consideration and further minimize the difference through the competition between a generator and discriminator, thereby greatly decreasing cross-domain distribution difference. Experimental results and comparison with several state-of-the-art methods verify GADM's superiority in image classification across domains.

Deep computational pathology in breast cancer.

Deep Learning (DL) algorithms are a set of techniques that exploit large and/or complex real-world datasets for cross-domain and cross-discipline prediction and classification tasks. DL architectures excel in computer vision tasks, and in particular image processing and interpretation. This has prompted a wave of disruptingly innovative applications in medical imaging, where DL strategies have the potential to vastly outperform human experts. This is particularly relevant in the context of histopathology, where whole slide imaging (WSI) of stained tissue in conjuction with DL algorithms for their interpretation, selection and cancer staging are beginning to play an ever increasing role in supporting human operators in visual assessments. This has the potential to reduce everyday workload as well as to increase precision and reproducibility across observers, centers, staining techniques and even pathologies. In this paper we introduce the most common DL architectures used in image analysis, with a focus on histopathological image analysis in general and in breast histology in particular. We briefly review how, state-of-art DL architectures compare to human performance on across a number of critical tasks such as mitotic count, tubules analysis and nuclear pleomorphism analysis. Also, the development of DL algorithms specialized to pathology images have been enormously fueled by a number of world-wide challenges based on large, multicentric image databases which are now publicly available. In turn, this has allowed most recent efforts to shift more and more towards semi-supervised learning methods, which provide greater flexibility and applicability. We also review all major repositories of manually labelled pathology images in breast cancer and provide an in-depth discussion of the challenges specific to training DL architectures to interpret WSI data, as well as a review of the state-of-the-art methods for interpretation of images generated from immunohistochemical analysis of breast lesions. We finally discuss the future challenges and opportunities which the adoption of DL paradigms is most likely to pose in the field of pathology for breast cancer detection, diagnosis, staging and prognosis. This review is intended as a comprehensive stepping stone into the field of modern computational pathology for a transdisciplinary readership across technical and medical disciplines.

360 degree view of cross-domain opinion classification: a survey

In the field of natural language processing and text mining, sentiment analysis (SA) has received huge attention from various researchers’ across the globe. By the prevalence of Web 2.0, user’s became more vigilant to share, promote and express themselves along with any issues or challenges that are being encountered on daily activities through the Internet (social media, micro-blogs, e-commerce, etc.) Expression and opinion are a complex sequence of acts that convey a huge volume of data that pose a challenge for computational researchers to decode. Over the period of time, researchers from various segments of public and private sectors are involved in the exploration of SA with an aim to understand the behavioral perspective of various stakeholders in society. Though the efforts to positively construct SA are successful, challenges still prevail for efficiency. This article presents an organized survey of SA (also known as opinion mining) along with methodologies or algorithms. The survey classifies SA into categories based on levels, tasks, and sub-task along with various techniques used for performing them. The survey explicitly focuses on different directions in which the research was explored in the area of cross-domain opinion classification. The article is concluded with an objective to present an exclusive and exhaustive analysis in the area of opinion mining containing approaches, datasets, languages, and applications used. The observations made are expected to support researches to get a greater understanding on emerging trends and state-of-the-art methods to be applied for future exploration.

Cross-domain few-shot classification through feature confusion

Few-shot classification has made great progress in recent years which aims to solve new tasks using the prior knowledge obtained from a set of similar tasks with few labeled examples. However, the performance of current methods drops when domain shift exists between training and testing samples. Considering the situation where target domain examples are hard to collect, it is necessary to focus on the cross-domain problem. In this paper, we propose a simple method which embeds the domain adaptation method into the few-shot problem under a novel cross-domain few-shot setting, and focus on supervised setting in the target domain where few labeled data available for domain adaptation. Extensive empirical evidence shows the effectiveness of our method.

Machine learning-assisted cross-domain prediction of ionic conductivity in sodium and lithium-based superionic conductors using facile descriptors

Solid state lithium- and sodium-ion batteries utilize solid ionically conducting compounds as electrolytes. However, the ionic conductivity of such materials tends to be lower than their liquid counterparts, necessitating research efforts into finding suitable alternatives. The process of electrolyte screening is often based on a mixture of domain expertise and trial-and-error, both of which are time and resource-intensive. In this work, we present a novel machine-learning based approach to predict the ionic conductivity of sodium and lithium-based SICON compounds. Using primarily theoretical elemental feature descriptors derivable from tabulated information on the unit cell and the atomic properties of the components of a target compound on a limited dataset of 70 NASICON-examples, we have designed a logistic regression-based model capable of distinguishing between poor and good superionic conductors with a validation accuracy of over 84%. Moreover, we demonstrate how such a system is capable of cross-domain classification on lithium-based examples with the same accuracy, despite being introduced to zero lithium-based compounds during training. Through a systematic permutation-based evaluation process, we reduced the number of considered features from 47 to 7, reduction of over 83%, while simultaneously improving model performance. The contributions of different electronic and structural features to overall ionic conductivity is also discussed, and contrasted with accepted theories in literature. Our results demonstrate the utility of such a facile tool in providing opportunities for initial screening of potential candidates as solid-state electrolytes through the use of existing data examples and simple tabulated or calculated features, reducing the time-to-market of such materials by helping to focus efforts on promising candidates. Given enough data utilizing suitable descriptors, high accurate cross-domain classifiers could be created for experimentalists, improving laboratory and computational efficiency.

DART: Domain-Adversarial Residual-Transfer networks for unsupervised cross-domain image classification

The accuracy of deep learning (e.g., convolutional neural networks) for an image classification task critically relies on the amount of labeled training data. Aiming to solve an image classification task on a new domain that lacks labeled data but gains access to cheaply available unlabeled data, unsupervised domain adaptation is a promising technique to boost the performance without incurring extra labeling cost, by assuming images from different domains share some invariant characteristics. In this paper, we propose a new unsupervised domain adaptation method named Domain-Adversarial Residual-Transfer (DART) learning of deep neural networks to tackle cross-domain image classification tasks. In contrast to the existing unsupervised domain adaption approaches, the proposed DART not only learns domain-invariant features via adversarial training, but also achieves robust domain-adaptive classification via a residual-transfer strategy, all in an end-to-end training framework. We evaluate the performance of the proposed method for cross-domain image classification tasks on several well-known benchmark data sets, in which our method clearly outperforms the state-of-the-art approaches.

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.

Multi-Group Transfer Learning on Multiple Latent Spaces for Text Classification

Transfer learning aims to leverage valuable information in one domain to promote the learning tasks in the other domain. Some recent studies indicated that the latent information, which has a close relationship with the high-level concepts, are more suitable for cross-domain text classification than learning raw features. To obtain more latent information existing in the latent feature space, some previous methods constructed multiple latent feature spaces. However, those methods ignored that the latent information of different latent spaces may lack the relevance for promoting the adaptability of transfer learning models, even may lead to negative knowledge transfer when there exists a glaring discrepancy among the different latent spaces. Additionally, since those methods learn the latent space distributions using a strategy of direct-promotion, their computational complexity increases exponentially as the number of latent spaces increases. To tackle this challenge, this paper proposes a Multiple Groups Transfer Learning (MGTL) method. MGTL first constructs multiple different latent feature spaces and then integrates the adjacent ones that have a similar latent feature dimension into one latent space group. Along this way, multiple latent space groups can be obtained. To enhance the relevance among these latent space groups, MGTL makes the adjacent groups contain one same latent space at least. Then, different groups will have more relevance than raw latent spaces. Second, MGTL utilizes an indirect-promotion strategy to connect different latent space groups. The computational complexity of MGTL increases linearly as the number of latent space groups increases and is superior to those multiple latent space methods based on direct-promotion. In addition, an iterative algorithm is proposed to solve the optimization problem. Finally, a set of systematic experiments demonstrate that MGTL outperforms all the compared existing methods.

Linear Discriminant Analysis via Pseudo Labels: A Unified Framework for Visual Domain Adaptation

This paper deals with the problem of visual domain adaptation in which source domain labeled data is available for training, but the target domain unlabeled data is available for testing. Many recent domain adaptation methods merely concentrate on extracting domain-invariant features via minimizing the distributional and geometrical divergence between domains simultaneously while ignoring within-class and between-class structure properties, especially for the target domain due to the unavailability of labeled data. We propose Linear Discriminant Analysis via Pseudo Labels (LDAPL), a unified framework for visual domain adaptation that can tackle these two issues together. LDAPL is to learn a domain-invariant features across both domains with preserving important properties such as minimizing the shift between domains both statistically and geometrically, retaining the original similarity of data samples, maximizing the target domain variance, and minimizing the within-class and maximizing the between-class properties of both domains. Specifically, LDAPL preserves the target domain discriminative information (or within-class and between-class properties) using pseudo labels, and these pseudo labels are refined until convergence. Extensive experiment on several visual cross-domain benchmarks, including Office+Caltech10 with all three types of features (such as Speeded Up Robust Features (SURF), Deep Convolutional Activation Feature (DeCAF <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">6</sub> ), and Visual Geometry Group-Fully Connected layer ( VGG-FC <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">6</sub> ) features), COIL20 (Columbia Object Image Library), digit, and PIE (Pose, Illumination, and Expression), LDAPL achieved average accuracies of 79.11%, 99.72%, 79.0%, and 84.50%, respectively. Comparative results on several visual cross-domain classification tasks verify that LDAPL can significantly outperform the state-of-the-art primitive and domain adaptation methods. Specifically, LDAPL gains over baseline Joint Geometrical and Statistical Alignment (JGSA) method with 6.6%, 5.3%, 6.3%, and 44.93% average accuracies on Office+Caltech10 (SURF, DeCAF <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">6</sub> , and VGG-FC6 ), COIL20, digit, and PIE, respectively.

Multisource Compensation Network for Remote Sensing Cross-Domain Scene Classification

Cross-domain scene classification refers to the scene classification task in which the training set (termed source domain) and the test set (termed target domain) come from different distributions. Various domain adaptation methods have been developed to reduce the distribution discrepancy between different domains. However, current domain adaptation methods assume that the source domain and target domain share the same categories. In reality, it is hard to find a source domain that can completely cover all the categories of target domain. In this article, we propose to use multiple complementary source domains to form the categories of target domain. A multisource compensation network (MSCN) is proposed to tackle these challenges: distribution discrepancy and category incompleteness. First, a pretrained convolutional neural network (CNN) is exploited to learn the feature representation for each domain. Second, a cross-domain alignment module is developed to reduce the domain shift between source and target domains. Domain shift is reduced by mapping the two domain features into a common feature space. Finally, a classifier complement module is proposed to align categories in multiple sources and learn a target classifier. Two cross-domain classification data sets are constructed using four heterogeneous remote sensing scene classification data sets. Extensive experiments are conducted on these datasets to validate the effectiveness of the proposed method. The proposed method can achieve 81.23% and 81.97% average accuracies on two-source-complementary data set and three-source-complementary data set, respectively.

A Multi-domain and Multi-modal Representation Disentangler for Cross-Domain Image Manipulation and Classification.

Learning interpretable data representation has been an active research topic in deep learning and computer vision. While representation disentanglement is an effective technique for addressing this task, existing works cannot easily handle the problems in which manipulating and recognizing data across multiple domains are desirable. In this paper, we present a unified network architecture of Multi-domain and Multi-modal Representation Disentangler (M2RD), with the goal of learning domain-invariant content representation with the associated domain-specific representation observed. By advancing adversarial learning and disentanglement techniques, the proposed model is able to perform continuous image manipulation across data domains with multiple modalities. More importantly, the resulting domain-invariant feature representation can be applied for unsupervised domain adaptation. Finally, our quantitative and qualitative results would confirm the effectiveness and robustness of the proposed model over state-of-the-art methods on the above tasks.

Class-specific Reconstruction Transfer Learning for Visual Recognition Across Domains.

Subspace learning and reconstruction have been widely explored in recent transfer learning work. Generally, a specially designed projection and reconstruction transfer functions bridging multiple domains for heterogeneous knowledge sharing are wanted. However, we argue that the existing subspace reconstruction based domain adaptation algorithms neglect the class prior, such that the learned transfer function is biased, especially when data scarcity of some class is encountered. Different from those previous methods, in this paper, we propose a novel class-wise reconstruction-based adaptation method called Class-specific Reconstruction Transfer Learning (CRTL), which optimizes a well modeled transfer loss function by fully exploiting intra-class dependency and inter-class independency. The merits of the CRTL are three-fold. 1) Using a class-specific reconstruction matrix to align the source domain with the target domain fully exploits the class prior in modeling the domain distribution consistency, which benefits the cross-domain classification. 2) Furthermore, to keep the intrinsic relationship between data and labels after feature augmentation, a projected Hilbert-Schmidt Independence Criterion (pHSIC), that measures the dependency between data and label, is first proposed in transfer learning community by mapping the data from raw space to RKHS. 3) In addition, by imposing low-rank and sparse constraints on the class-specific reconstruction coefficient matrix, the global and local data structure that contributes to domain correlation can be effectively preserved. Extensive experiments on challenging benchmark datasets demonstrate the superiority of the proposed method over state-of-the-art representation-based domain adaptation methods. The demo code is available in https://github.com/wangshanshanCQU/CRTL.

Extreme Learning Machine-Based Heterogeneous Domain Adaptation for Classification of Hyperspectral Images

An extreme learning machine (ELM)-based heterogeneous domain adaptation (HDA) algorithm is proposed for the classification of remote sensing images. In the adaptive ELM network, one hidden layer is used for the source data to provide the random features, whereas two hidden layers are set for target data to produce the random features as well as a transformation matrix. DA is achieved by constraining both the source data and the transformed target data to share the same output weights. Moreover, manifold regularization is adopted to preserve the local geometry of unlabeled target data. The proposed ELM-based HDA (EHDA) method is applied to cross-domain classification of remote sensing images, and the experimental results using multisensor remote sensing images demonstrate the effectiveness of the proposed approach.

Unsupervised Manifold Alignment for Cross-Domain Classification of Remote Sensing Images

The original manifold alignment (MA) approach is for semisupervised domain adaptation. Since the target prior information is difficult to obtain, we conduct it in an unsupervised manner, resulting in an unsupervised MA (UMA) method. This approach utilizes the probabilistic prediction results of target data to construct the cross-domain similarity matrix, which characterizes the relationships between domains and is used for alignment. Due to the spectral drift, the prediction results may not be accurate, and thus affect the alignment. We employed spatial filtering and overall centroid alignment method as two preprocessing strategies to improve the prediction results. Furthermore, per-class maximum mean discrepancy (MMD) constraint is introduced to the UMA to further improve the alignment performance. The proposed UMA_MMD algorithm is applied for the classification of remote sensing images, and the experimental results using hyperion multitemporal remote sensing images demonstrated the effectiveness of the proposed approach.

Classifier-Constrained Deep Adversarial Domain Adaptation for Cross-Domain Semisupervised Classification in Remote Sensing Images

This letter presents a classifier-constrained deep adversarial domain adaptation (CDADA) method for cross-domain semisupervised classification in remote sensing (RS) images. A deep convolutional neural network (DCNN) is used to build feature representations to describe the semantic content of scenes before the adaptation process. Then, adversarial domain adaptation is used to align the feature distribution of the source and the target. Specifically, two different land-cover classifiers are used as a discriminator to consider land-cover decision boundaries between classes and increase their distance to separate them from the original land-cover class boundaries. The generator then creates robust transferable features far from the original land-cover class boundaries under the classifier constraint. The experimental results of six scenarios built from three benchmark RS scene data sets (AID, Merced, and RSI-CB data sets) are reported and discussed.

Multi-representation adaptation network for cross-domain image classification

In image classification, it is often expensive and time-consuming to acquire sufficient labels. To solve this problem, domain adaptation often provides an attractive option given a large amount of labeled data from a similar nature but different domains. Existing approaches mainly align the distributions of representations extracted by a single structure and the representations may only contain partial information, e.g., only contain part of the saturation, brightness, and hue information. Along this line, we propose Multi-Representation Adaptation which can dramatically improve the classification accuracy for cross-domain image classification and specially aims to align the distributions of multiple representations extracted by a hybrid structure named Inception Adaptation Module (IAM). Based on this, we present Multi-Representation Adaptation Network (MRAN) to accomplish the cross-domain image classification task via multi-representation alignment which can capture the information from different aspects. In addition, we extend Maximum Mean Discrepancy (MMD) to compute the adaptation loss. Our approach can be easily implemented by extending most feed-forward models with IAM, and the network can be trained efficiently via back-propagation. Experiments conducted on three benchmark image datasets demonstrate the effectiveness of MRAN.

Aligning Domain-Specific Distribution and Classifier for Cross-Domain Classification from Multiple Sources

While Unsupervised Domain Adaptation (UDA) algorithms, i.e., there are only labeled data from source domains, have been actively studied in recent years, most algorithms and theoretical results focus on Single-source Unsupervised Domain Adaptation (SUDA). However, in the practical scenario, labeled data can be typically collected from multiple diverse sources, and they might be different not only from the target domain but also from each other. Thus, domain adapters from multiple sources should not be modeled in the same way. Recent deep learning based Multi-source Unsupervised Domain Adaptation (MUDA) algorithms focus on extracting common domain-invariant representations for all domains by aligning distribution of all pairs of source and target domains in a common feature space. However, it is often very hard to extract the same domain-invariant representations for all domains in MUDA. In addition, these methods match distributions without considering domain-specific decision boundaries between classes. To solve these problems, we propose a new framework with two alignment stages for MUDA which not only respectively aligns the distributions of each pair of source and target domains in multiple specific feature spaces, but also aligns the outputs of classifiers by utilizing the domainspecific decision boundaries. Extensive experiments demonstrate that our method can achieve remarkable results on popular benchmark datasets for image classification.