Unlabeled Target Data Research Articles

Retrieve-and-Edit Domain Adaptation for End2End Aspect Based Sentiment Analysis

End-to-end aspect based sentiment analysis (E2E-ABSA) aims to jointly extract aspect terms and predict aspect-level sentiment for opinion reviews. Though supervised methods show effectiveness for E2E-ABSA tasks, the annotation cost is extremely high due to the necessity of fine-grained labels. Recent attempts alleviate this problem using the domain adaptation technique to transfer the word-level common knowledge across domains. However, the biggest issue in domain adaptation, i.e., how to transfer the domain-specific words like <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">pizza</i> and <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">delicious</i> in the source “Restaurant” to the target “Laptop” domain, has not been resolved. In this paper, we propose a novel domain adaptation method to address this issue by enhancing the transferability of domain-specific source words in a retrieve-and-edit way. Specifically, for all source words, we first retrieve the transferable prototypes from unlabeled target data via their syntactic and semantic roles. We then edit the source words to enhance their transferability by absorbing the knowledge carried in prototypes. Finally, we design an end-to-end framework to jointly accomplish cross-domain aspect term extraction and aspect-level sentiment classification. We conduct extensive experiments on four real-world datasets. The results prove that, by introducing transferable prototypes, our method significantly outperforms the state-of-the-art methods, achieving an absolute 3.95% F1 increase over the best baseline.

Non-Parametric Bayesian Subspace Models for Acoustic Unit Discovery

This work investigates subspace non-parametric models for the task of learning a set of acoustic units from unlabeled speech recordings. We constrain the base-measure of a Dirichlet-Process mixture with a phonetic subspace---estimated from other source languages---to build an \emph{educated prior}, thereby forcing the learned acoustic units to resemble phones of known source languages. Two types of models are proposed: (i) the Subspace HMM (SHMM) which assumes that the phonetic subspace is the same for every language, (ii) the Hierarchical-Subspace HMM (H-SHMM) which relaxes this assumption and allows to have a language-specific subspace estimated on the unlabeled target data. These models are applied on 3 languages: English, Yoruba and Mboshi and they are compared with various competitive acoustic units discovery baselines. Experimental results show that both subspace models outperform other systems in terms of clustering quality and segmentation accuracy. Moreover, we observe that the H-SHMM provides results superior to the SHMM supporting the idea that language-specific priors are preferable to language-agnostic priors for acoustic unit discovery.

ACT: Semi-supervised Domain-adaptive Medical Image Segmentation with Asymmetric Co-Training.

Unsupervised domain adaptation (UDA) has been vastly explored to alleviate domain shifts between source and target domains, by applying a well-performed model in an unlabeled target domain via supervision of a labeled source domain. Recent literature, however, has indicated that the performance is still far from satisfactory in the presence of significant domain shifts. Nonetheless, delineating a few target samples is usually manageable and particularly worthwhile, due to the substantial performance gain. Inspired by this, we aim to develop semi-supervised domain adaptation (SSDA) for medical image segmentation, which is largely underexplored. We, thus, propose to exploit both labeled source and target domain data, in addition to unlabeled target data in a unified manner. Specifically, we present a novel asymmetric co-training (ACT) framework to integrate these subsets and avoid the domination of the source domain data. Following a divide-and-conquer strategy, we explicitly decouple the label supervisions in SSDA into two asymmetric sub-tasks, including semi-supervised learning (SSL) and UDA, and leverage different knowledge from two segmentors to take into account the distinction between the source and target label supervisions. The knowledge learned in the two modules is then adaptively integrated with ACT, by iteratively teaching each other, based on the confidence-aware pseudo-label. In addition, pseudo label noise is well-controlled with an exponential MixUp decay scheme for smooth propagation. Experiments on cross-modality brain tumor MRI segmentation tasks using the BraTS18 database showed, even with limited labeled target samples, ACT yielded marked improvements over UDA and state-of-the-art SSDA methods and approached an "upper bound" of supervised joint training.

Confident Learning-Based Domain Adaptation for Hyperspectral Image Classification

Cross-domain hyperspectral image classification is one of the major challenges in remote sensing, especially for target domain data without labels. Recently, deep learning approaches have demonstrated effectiveness in domain adaptation. However, most of them leverage unlabeled target data only from a statistical perspective but neglect the analysis at the instance level. For better statistical alignment, existing approaches employ the entire unevaluated target data in an unsupervised manner, which may introduce noise and limit the discriminability of the neural networks. In this article, we propose confident learning-based domain adaptation (CLDA) to address the problem from a new perspective of data manipulation. To this end, a novel framework is presented to combine domain adaptation with confident learning (CL), where the former reduces the interdomain discrepancy and generates pseudo-labels for the target instances, from which the latter selects high-confidence target samples. Specifically, the confident learning part evaluates the confidence of each pseudo-labeled target sample based on the assigned labels and the predicted probabilities. Then, high-confidence target samples are selected as training data to increase the discriminative capacity of the neural networks. In addition, the domain adaptation part and the confident learning part are trained alternately to progressively increase the proportion of high-confidence labels in the target domain, thus further improving the accuracy of classification. Experimental results on four datasets demonstrate that the proposed CLDA method outperforms the state-of-the-art domain adaptation approaches. Our source code is available at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/Li-ZK/CLDA-2022</uri> .

Generalized Meta-FDMixup: Cross-Domain Few-Shot Learning Guided by Labeled Target Data.

The vanilla Few-shot Learning (FSL) learns to build a classifier for a new concept from one or very few target examples, with the general assumption that source and target classes are sampled from the same domain. Recently, the task of Cross-Domain Few-Shot Learning (CD-FSL) aims at tackling the FSL where there is a huge domain shift between the source and target datasets. Extensive efforts on CD-FSL have been made via either directly extending the meta-learning paradigm of vanilla FSL methods, or employing massive unlabeled target data to help learn models. In this paper, we notice that in the CD-FSL task, the few labeled target images have never been explicitly leveraged to inform the model in the training stage. However, such a labeled target example set is very important to bridge the huge domain gap. Critically, this paper advocates a more practical training scenario for CD-FSL. And our key insight is to utilize a few labeled target data to guide the learning of the CD-FSL model. Technically, we propose a novel Generalized Meta-learning based Feature-Disentangled Mixup network, namely GMeta-FDMixup. We make three key contributions of utilizing GMeta-FDMixup to address CD-FSL. Firstly, we present two mixup modules - mixup-P and mixup-M that help facilitate utilizing the unbalanced and disjoint source and target datasets. These two novel modules enable diverse image generation for training the model on the source domain. Secondly, to narrow the domain gap explicitly, we contribute a novel feature disentanglement module that learns to decouple the domain-irrelevant and domain-specific features. By stripping the domain-specific features, we alleviate the negative effects caused by the domain inductive bias. Finally, we repurpose a new contrastive learning module, dubbed ConL. ConL prevents the model from only capturing category-related features via introducing contrastive loss. Thus, the generalization ability on novel categories is improved. Extensive experimental results on two benchmarks show the superiority of our setting and the effectiveness of our method. Code and models will be released.

Revealing Task-Relevant Model Memorization for Source-Protected Unsupervised Domain Adaptation

Source-data-free unsupervised domain adaptation (SF-UDA) is an approach to improve model performance in the target domain without accessing the source data. Some SF-UDA methods have been proposed and achieved promising results using the information from source-model parameters. However, current research on information security confirms the ability of a well-trained model to memorize its training data. Therefore, SF-UDA methods that access model parameters remain at risk of privacy disclosure. This paper introduces a new topic of source-protected UDA (SP-UDA) that adapts the source model to the target domain while protecting the source-domain data and model privacy. In SP-UDA, only a black-box source model and a set of unlabeled target data are available for domain adaptation. We consider SP-UDA from a new perspective of model memorization revelation. A Source-Protected Generative Model (SPGM) is developed to reveal task-relevant memorization from the source model. SPGM directly distills the inverse process of the source model without access to source-model parameters to meet the privacy protection objective in SP-UDA. The SPGM is learned under the supervision of a newly designed metric named privacy-protected transfer (PPT). The PPT metric measures the transferability and desensitization of the generated data to encourage the SPGM to extract task-relevant information rather than the unintended memorization. A set of desensitized pseudo data is then generated as substitutes for the real source data in UDA. The performance of the proposed method has been validated in four cross-dataset recognition applications with encouraging results.

Deep reconstruction-recoding network for unsupervised domain adaptation and multi-center generalization in colonoscopy polyp detection

Background and objective: Currently, the best performing methods in colonoscopy polyp detection are primarily based on deep neural networks (DNNs), which are usually trained on large amounts of labeled data. However, different hospitals use different endoscope models and set different imaging parameters, which causes the collected endoscopic images and videos to vary greatly in style. There may be variations in the color space, brightness, contrast, and resolution, and there are also differences between white light endoscopy (WLE) and narrow band image endoscopy (NBIE). We call these variations the domain shift. The DNN performance may decrease when the training data and the testing data come from different hospitals or different endoscope models. Additionally, it is quite difficult to collect enough new labeled data and retrain a new DNN model before deploying that DNN to a new hospital or endoscope model.Methods: To solve this problem, we propose a domain adaptation model called Deep Reconstruction-Recoding Network (DRRN), which jointly learns a shared encoding representation for two tasks: i) a supervised object detection network for labeled source data, and ii) an unsupervised reconstruction-recoding network for unlabeled target data. Through the DRRN, the object detection network's encoder not only learns the features from the labeled source domain, but also encodes useful information from the unlabeled target domain. Therefore, the distribution difference of the two domains’ feature spaces can be reduced.Results: We evaluate the performance of the DRRN on a series of cross-domain datasets. Compared with training the polyp detection network using only source data, the performance of the DRRN on the target domain is improved. Through feature statistics and visualization, it is demonstrated that the DRRN can learn the common distribution and feature invariance of the two domains. The distribution difference between the feature spaces of the two domains can be reduced.Conclusion: The DRRN can improve cross-domain polyp detection. With the DRRN, the generalization performance of the DNN-based polyp detection model can be improved without additional labeled data. This improvement allows the polyp detection model to be easily transferred to datasets from different hospitals or different endoscope models.

Domain Adaptation for Structural Fault Detection under Model Uncertainty

In the last decade, the interest in machine learning (ML) has grown significantly within the structural health monitoring (SHM) community. Traditional supervised ML approaches for detecting faults assume that the training and test data come from similar distributions. However, real-world applications, where an ML model is trained, for example, on numerical simulation data and tested on experimental data, are deemed to fail in detecting the damage. The deterioration in the prediction performance is mainly related to the fact that the numerical and experimental data are collected under different conditions and they do not share the same underlying features. This paper proposes a domain adaptation approach for ML-based damage detection and localization problems where the classifier has access to the labeled training (source) and unlabeled test (target) data, but the source and target domains are statistically different. The proposed domain adaptation method seeks to form a feature space that is capable of representing both source and target domains by implementing a domain-adversarial neural network. This neural network uses H-divergence criteria to minimize the discrepancy between the source and target domain in a latent feature space. To evaluate the performance, we present two case studies where we design a neural network model for classifying the health condition of a variety of systems. The effectiveness of the domain adaptation is shown by computing the classification accuracy of the unlabeled target data with and without domain adaptation. Furthermore, the performance gain of the domain adaptation over a well-known transfer knowledge approach called Transfer Component Analysis is also demonstrated. Overall, the results demonstrate that the domain adaption is a valid approach for damage detection applications where access to labeled experimental data is limited.

Synthetic Source Universal Domain Adaptation through Contrastive Learning.

Universal domain adaptation (UDA) is a crucial research topic for efficient deep learning model training using data from various imaging sensors. However, its development is affected by unlabeled target data. Moreover, the nonexistence of prior knowledge of the source and target domain makes it more challenging for UDA to train models. I hypothesize that the degradation of trained models in the target domain is caused by the lack of direct training loss to improve the discriminative power of the target domain data. As a result, the target data adapted to the source representations is biased toward the source domain. I found that the degradation was more pronounced when I used synthetic data for the source domain and real data for the target domain. In this paper, I propose a UDA method with target domain contrastive learning. The proposed method enables models to leverage synthetic data for the source domain and train the discriminativeness of target features in an unsupervised manner. In addition, the target domain feature extraction network is shared with the source domain classification task, preventing unnecessary computational growth. Extensive experimental results on VisDa-2017 and MNIST to SVHN demonstrated that the proposed method significantly outperforms the baseline by 2.7% and 5.1%, respectively.

Towards Cross-Environment Human Activity Recognition Based on Radar Without Source Data

Radar-based human activity recognition (HAR) finds various applications like assisted living and driver behavior monitoring. As radar data are heavily environment-dependent, it is becoming increasingly important to develop a transfer learning mechanism that enables a radar-based HAR system with desirable cross-environment adaptation feasibility. This paper concerns the issue of how radar-based HAR system can adapt to a new environment without source data. To this end, we devote to using the source hypothesis transfer learning architecture to build such an environment adaptation mechanism towards cross-environment radar-based HAR. In doing this, it is a challenging task to develop a reliable self-supervised labeling strategy for generating pseudo labels associated with the unlabeled target data, which is crucial to facilitate the learning of a target-specific feature extractor being responsible for environment adaptation. This paper presents the neighbor-aggregating-based labeling method and incorporates it with the existing clustering-based labeling method to perform the self-supervised labeling task. The logic behind our approach is that the above two labeling methods are complementary to each other in terms of making use of both local and global structures of adaptation data to supervise the labeling task. The coordination of both labeling methods is motivated to be implemented in the weighted combination form, which contributes to improving the reliability of generated labels. Experimental results on a public HAR dataset based on the frequency modulated continuous wave (FMCW) radar demonstrate the effectiveness of our approach.

Cross-Modality Transfer Learning for Image-Text Information Management

In the past decades, information from all kinds of data has been on a rapid increase. With state-of-the-art performance, machine learning algorithms have been beneficial for information management. However, insufficient supervised training data is still an adversity in many real-world applications. Therefore, transfer learning (TF) was proposed to address this issue. This article studies a not well investigated but important TL problem termed cross-modality transfer learning (CMTL). This topic is closely related to distant domain transfer learning (DDTL) and negative transfer. In general, conventional TL disciplines assume that the source domain and the target domain are in the same modality. DDTL aims to make efficient transfers even when the domains or the tasks are entirely different. As an extension of DDTL, CMTL aims to make efficient transfers between two different data modalities, such as from image to text. As the main focus of this study, we aim to improve the performance of image classification by transferring knowledge from text data. Previously, a few CMTL algorithms were proposed to deal with image classification problems. However, most existing algorithms are very task specific, and they are unstable on convergence. There are four main contributions in this study. First, we propose a novel heterogeneous CMTL algorithm, which requires only a tiny set of unlabeled target data and labeled source data with associate text tags. Second, we introduce a latent semantic information extraction method to connect the information learned from the image data and the text data. Third, the proposed method can effectively handle the information transfer across different modalities (text-image). Fourth, we examined our algorithm on a public dataset, Office-31. It has achieved up to 5% higher classification accuracy than “non-transfer” algorithms and up to 9% higher than existing CMTL algorithms.

Learning Domain Invariant Representation via Self-Rugularization

Unsupervised domain adaptation often gives impressive solutions to handle domain shift of data. Most of current approaches assume that unlabeled target data to train is abundant. This assumption is not always true in practices. To tackle this issue, we propose a general solution to solve the domain gap minimization problem without any target data. Our method consists of two regularization steps. The first step is a pixel regularization by arbitrary style transfer. Recently, some methods bring style transfer algorithms to domain adaptation and domain generalization process. They use style transfer algorithms to remove texture bias in source domain data. We also use style transfer algorithms for removing texture bias, but our method depends on neither domain adaptation nor domain generalization paradigm. The second regularization step is a feature regularization by feature alignment. Adding a feature alignment loss term to the model loss, the model learns domain invariant representation more efficiently. We evaluate our regularization methods from several experiments both on small dataset and large dataset. From the experiments, we show that our model can learn domain invariant representation as much as unsupervised domain adaptation methods.

Co‐teaching based pseudo label refinery for cross‐domain object detection

Object detection is one of the main tasks in computer vision and has made great progress in recent years. However, the performance of target detectors is significantly dropped by the differences between existing datasets and application scenarios, leading to the so-called domain shift problem. To address such an issue, a novel co-teaching based pseudo label refinery framework for cross-domain object detection is developed, which cooperates with two models to select data from target domain for each other. This strategy can effectively purify the predicted pseudo labels and resist noisy labels. Specifically, the framework consists of two encoders (i.e. structure encoder and global encoder), two classifiers and one discriminator, in which structure encoder is used to extract structural features that are not disturbed by colour, and the global encoder is used to extract the complete discriminant features. The two encoders are each followed by a classifier. In training, the structure and global encoder with labelled source samples are first trained, so that it has the initial recognition ability. Then the samples assigned are used with pseudo labels by the classifier following the structure encoder to fine-tune the global encoder which pre-trained on the labelled source domain and obtain the refined labels for the target data. With the refined labels, the structure encoder is further optimised on the target domain. During this process, the proposal is to cross use the two classifiers to promote the mutual transfer of complementary capabilities of the two encoders. Moreover, a novel residual channel attention block (RCA) embedded with salient features is designed to pay more attention to the target regions. Extensive experiments demonstrate that the developed framework can generate clean labels for unlabelled target data and boost the performance of cross domain object detection. The code is available at http://www.msp-lab.cn:1436/msp/cbplr-master.

ScAdapt: virtual adversarial domain adaptation network for single cell RNA-seq data classification across platforms and species.

In single cell analyses, cell types are conventionally identified based on expressions of known marker genes, whose identifications are time-consuming and irreproducible. To solve this issue, many supervised approaches have been developed to identify cell types based on the rapid accumulation of public datasets. However, these approaches are sensitive to batch effects or biological variations since the data distributions are different in cross-platforms or species predictions. In this study, we developed scAdapt, a virtual adversarial domain adaptation network, to transfer cell labels between datasets with batch effects. scAdapt used both the labeled source and unlabeled target data to train an enhanced classifier and aligned the labeled source centroids and pseudo-labeled target centroids to generate a joint embedding. The scAdapt was demonstrated to outperform existing methods for classification in simulated, cross-platforms, cross-species, spatial transcriptomic and COVID-19 immune datasets. Further quantitative evaluations and visualizations for the aligned embeddings confirm the superiority in cell mixing and the ability to preserve discriminative cluster structure present in the original datasets.

Unsupervised multi-source domain adaptation with no observable source data.

Given trained models from multiple source domains, how can we predict the labels of unlabeled data in a target domain? Unsupervised multi-source domain adaptation (UMDA) aims for predicting the labels of unlabeled target data by transferring the knowledge of multiple source domains. UMDA is a crucial problem in many real-world scenarios where no labeled target data are available. Previous approaches in UMDA assume that data are observable over all domains. However, source data are not easily accessible due to privacy or confidentiality issues in a lot of practical scenarios, although classifiers learned in source domains are readily available. In this work, we target data-free UMDA where source data are not observable at all, a novel problem that has not been studied before despite being very realistic and crucial. To solve data-free UMDA, we propose DEMS (Data-free Exploitation of Multiple Sources), a novel architecture that adapts target data to source domains without exploiting any source data, and estimates the target labels by exploiting pre-trained source classifiers. Extensive experiments for data-free UMDA on real-world datasets show that DEMS provides the state-of-the-art accuracy which is up to 27.5% point higher than that of the best baseline.

Deep Ladder-Suppression Network for Unsupervised Domain Adaptation.

Unsupervised domain adaptation (UDA) aims at learning a classifier for an unlabeled target domain by transferring knowledge from a labeled source domain with a related but different distribution. Most existing approaches learn domain-invariant features by adapting the entire information of the images. However, forcing adaptation of domain-specific variations undermines the effectiveness of the learned features. To address this problem, we propose a novel, yet elegant module, called the deep ladder-suppression network (DLSN), which is designed to better learn the cross-domain shared content by suppressing domain-specific variations. Our proposed DLSN is an autoencoder with lateral connections from the encoder to the decoder. By this design, the domain-specific details, which are only necessary for reconstructing the unlabeled target data, are directly fed to the decoder to complete the reconstruction task, relieving the pressure of learning domain-specific variations at the later layers of the shared encoder. As a result, DLSN allows the shared encoder to focus on learning cross-domain shared content and ignores the domain-specific variations. Notably, the proposed DLSN can be used as a standard module to be integrated with various existing UDA frameworks to further boost performance. Without whistles and bells, extensive experimental results on four gold-standard domain adaptation datasets, for example: 1) Digits; 2) Office31; 3) Office-Home; and 4) VisDA-C, demonstrate that the proposed DLSN can consistently and significantly improve the performance of various popular UDA frameworks.

Causal Mechanism Transfer Network for Time Series Domain Adaptation in Mechanical Systems

Data-driven models are becoming essential parts in modern mechanical systems, commonly used to capture the behavior of various equipment and varying environmental characteristics. Despite the advantages of these data-driven models on excellent adaptivity to high dynamics and aging equipment, they are usually hungry for massive labels, mostly contributed by human engineers at a high cost. Fortunately, domain adaptation enhances the model generalization by utilizing the labeled source data and the unlabeled target data. However, the mainstream domain adaptation methods cannot achieve ideal performance on time series data, since they assume that the conditional distributions are equal. This assumption works well in the static data but is inapplicable for the time series data. Even the first-order Markov dependence assumption requires the dependence between any two consecutive time steps. In this article, we assume that the causal mechanism is invariant and present our Causal Mechanism Transfer Network (CMTN) for time series domain adaptation. By capturing causal mechanisms of time series data, CMTN allows the data-driven models to exploit existing data and labels from similar systems, such that the resulting model on a new system is highly reliable even with limited data. We report our empirical results and lessons learned from two real-world case studies, on chiller plant energy optimization and boiler fault detection, which outperform the existing state-of-the-art method.

Adaptive data augmentation for supervised learning over missing data

Real-world data is dirty, which causes serious problems in (supervised) machine learning (ML). The widely used practice in such scenario is to first repair the labeled source (a.k.a. train) data using rule-, statistical- or ML-based methods and then use the "repaired" source to train an ML model. During production, unlabeled target (a.k.a. test) data will also be repaired, and is then fed in the trained ML model for prediction. However, this process often causes a performance degradation when the source and target datasets are dirty with different noise patterns , which is common in practice. In this paper, we propose an adaptive data augmentation approach, for handling missing data in supervised ML. The approach extracts noise patterns from target data, and adapts the source data with the extracted target noise patterns while still preserving supervision signals in the source. Then, it patches the ML model by retraining it on the adapted data, in order to better serve the target. To effectively support adaptive data augmentation, we propose a novel generative adversarial network (GAN) based framework, called DAGAN, which works in an unsupervised fashion. DAGAN consists of two connected GAN networks. The first GAN learns the noise pattern from the target, for target mask generation. The second GAN uses the learned target mask to augment the source data, for source data adaptation. The augmented source data is used to retrain the ML model. Extensive experiments show that our method significantly improves the ML model performance and is more robust than the state-of-the-art missing data imputation solutions for handling datasets with different missing value patterns.

Similarity-based meta-learning network with adversarial domain adaptation for cross-domain fault identification

With wide applications of intelligent methods in mechanical fault diagnosis, satisfactory results have been achieved. However, complicated and diverse practical working conditions would significantly reduce the performance of the diagnostic model that works well in the laboratory, i.e. domain shift occurs. To address the problem, this paper proposed a novel similarity-based meta-learning network with adversarial domain adaptation for cross-domain fault identification. The proposed domain-adversarial similarity-based meta-learning network (DASMN) consists of three modules: a feature encoder, a classifier and a domain discriminator. First, the encoder and the classifier implement the similarity-based meta-learning algorithm, in while the good generalization ability for unseen tasks is obtained. Then, adversarial domain adaptation is conducted by minimizing and maximizing the domain-discriminative error adversarially, which takes unlabeled source data and target data as inputs. The effectiveness of DASMN is evaluated by multiple cross-domain cases using three bearing vibration datasets and is compared with five well-established methods. Experimental results demonstrate the availability and outstanding generalization ability of the proposed method for cross-domain fault identification.

Label propagation with multi-stage inference for visual domain adaptation

Domain adaptation aims at leveraging rich label information in source domain and predicting on a related but different target domain, which makes significant progress in visual recognition. Most existing methods attempt to align domains but have limited considerations on the preservation of data structure, especially for unlabeled target data. In this paper, we present a novel solution for unsupervised domain adaptation based on label propagation. Specifically, we construct the affinity graph separately to capture both within-domain and cross-domain property. Inspired by the overlap exists between two domains, we propose multi-stage inference. Samples shared by two domains are expected to have more reliable labels, then they are picked to assist further predictions. Extensive experiments on three real-world datasets demonstrate that our method is comparable or superior to existing methods, we also provide some discussion about the interpretation of multi-stage inference.