Unsupervised New-set Domain Adaptation with Self-supervised Knowledge

Unsupervised Domain Adaptation (UDA) aims to use the source domain with large amounts of labeled data to help the learning of the target domain without any label information. In UDA, the source and target domains are usually assumed to have different data distributions but share the same class label space. Nevertheless, in real-world open learning scenarios, label spaces are highly likely to be different across domains. In extreme cases, the domains share no common classes, i.e., all classes in the target domain are new classes. In such a case, direct transferring the class-discriminative knowledge from the source domain may impair the performance in the target domain and lead to negative transfer. For this reason, this paper proposes unsupervised new-set domain adaptation with self-supervised knowledge (SUNDA) to transfer the sample contrastive knowledge from the source domain, and use self-supervised knowledge from the target domain to guide the knowledge transfer. Specifically, the initial features of the source and target domains are learned by self-supervised learning, and some network parameters are frozen to preserve target domain information. Sample contrastive knowledge from the source domain is then transferred to the target domain to assist the learning of class-discriminative features in the target domain. Moreover, graph-based self-supervised classification loss is adopted to handle the problem of target domain classification with no inter-domain common classes. SUNDA is evaluated on tasks of cross-domain transfer for handwritten digits without any common class and cross-race transfer for face data without any common class. The experiments show that SUNDA outperforms UDA, unsupervised clustering, and new class discovery methods in learning performance.     

Active domain adaptation with noisy labels for multimedia analysis

Supervised learning methods require sufficient labeled examples to learn a good model for classification or regression. However, available labeled data are insufficient in many applications. Active learning (AL) and domain adaptation (DA) are two strategies to minimize the required amount of labeled data for model training. AL requires the domain expert to label a small number of highly informative examples to facilitate classification, while DA involves tuning the source domain knowledge for classification on the target domain. In this paper, we demonstrate how AL can efficiently minimize the required amount of labeled data for DA. Since the source and target domains usually have different distributions, it is possible that the domain expert may not have sufficient knowledge to answer each query correctly. We exploit our active DA framework to handle incorrect labels provided by domain experts. Experiments with multimedia data demonstrate the efficiency of our proposed framework for active DA with noisy labels.     

基于自监督知识的无监督新集域适应学习

无监督域适应(unsupervised domain adaptation,UDA)旨在利用带大量标注数据的源域帮助无任何标注信息的目标域学习.在UDA中,通常假设源域和目标域间的数据分布不同,但共享相同的类标签空间.但在真实开放学习场景中,域间的标签空间很可能存在差异.在极端情形下,域间的类别不存在交集,即目标域中类...     

基于动态策略的多源迁移学习数据流分类研究

为解决数据流分类中的概念漂移和噪声问题,提出一种基于样本确定性的多源迁移学习方法。该方法存储多源领域上由训练得到的分类器,求出各源领域分类器对目标领域数据块中每个样本的类别后验概率和样本确定性值。在此基础上,将样本确定性值满足当前阈值限制的源领域分类器与目标领域分类器进行在线集成,从而将多个源领域的知识迁移到目标领域。实验结果表明,该方法能够有效消除噪声数据流给不确定分类器带来的不利影响,与基于准确率选择集成的多源迁移学习方法相比,具有更高的分类准确率和抗噪稳定性。     

Linear semi-supervised projection clustering by transferred centroid regularization

We propose a novel method, called Semi-supervised Projection Clustering in Transfer Learning (SPCTL), where multiple source domains and one target domain are assumed. Traditional semi-supervised projection clustering methods hold the assumption that the data and pairwise constraints are all drawn from the same domain. However, many related data sets with different distributions are available in real applications. The traditional methods thus can not be directly extended to such a scenario. One major challenging issue is how to exploit constraint knowledge from multiple source domains and transfer it to the target domain where all the data are unlabeled. To handle this difficulty, we are motivated to construct a common subspace where the difference in distributions among domains can be reduced. We also invent a transferred centroid regularization, which acts as a bridge to transfer the constraint knowledge to the target domain, to formulate this geometric structure formed by the centroids from different domains. Extensive experiments on both synthetic and benchmark data sets show the effectiveness of our method.     

Online transfer learning by leveraging multiple source domains

Transfer learning aims to enhance performance in a target domain by exploiting useful information from auxiliary or source domains when the labeled data in the target domain are insufficient or difficult to acquire. In some real-world applications, the data of source domain are provided in advance, but the data of target domain may arrive in a stream fashion. This kind of problem is known as online transfer learning. In practice, there can be several source domains that are related to the target domain. The performance of online transfer learning is highly associated with selected source domains, and simply combining the source domains may lead to unsatisfactory performance. In this paper, we seek to promote classification performance in a target domain by leveraging labeled data from multiple source domains in online setting. To achieve this, we propose a new online transfer learning algorithm that merges and leverages the classifiers of the source and target domain with an ensemble method. The mistake bound of the proposed algorithm is analyzed, and the comprehensive experiments on three real-world data sets illustrate that our algorithm outperforms the compared baseline algorithms.     

基于深度迁移学习的网络入侵检测

为解决网络入侵检测问题,提高检测准确率和降低误报率,提出一种基于深度迁移学习的网络入侵检测方法,该方法使用非监督学习的深度自编码器来进行迁移学习,实现网络的入侵检测。首先对深度迁移学习问题进行建模,然后对深度模型进行迁移学习。迁移学习框架由嵌入层和标签层实现编/解码,编码和解码权重由源域和目标域共享,用于知识的迁移。嵌入层中,通过最小化域之间的嵌入实例的KL散度来强制源域和目标域数据的分布相似;在标签编码层中,使用softmax回归模型对源域的标签信息进行编码分类。实验结果表明,该方法能够实现网络入侵检测,且性能优于其他入侵检测方法。     

基于dropout正则化的半监督域自适应方法

针对机器学习中训练样本和测试样本概率分布不一致的问题,提出了一种基于dropout正则化的半监督域自适应方法来实现将神经网络的特征表示从标签丰富的源域转移到无标签的目标域。此方法从半监督学习的角度出发,在源域数据中添加少量带标签的目标域数据,使得神经网络在学习到源域数据特征分布的同时也能学习到目标域数据的特征分布。由于有了先验知识的指导,即使没有丰富的标签信息,神经网络依然可以很好地拟合目标域数据。实验结果表明,此算法在几种典型的数字数据集SVHN、MNIST和USPS的域自适应任务上的性能优于现有的其他算法,并且在涵盖广泛自然类别的真实数据集CIFAR-10和STL-10的域自适应任务上有较好的鲁棒性。     

Multi-source unsupervised domain adaptation for object detection

Domain adaptation for object detection has been extensively studied in recent years. Most existing approaches focus on single-source unsupervised domain adaptive object detection. However, a more practical scenario is that the labeled source data is collected from multiple domains with different feature distributions. The conventional approaches do not work very well since multiple domain gaps exist. We propose a Multi-source domain Knowledge Transfer (MKT) method to handle this situation. First, the low-level features from multiple domains are aligned by learning a shallow feature extraction network. Then, the high-level features from each pair of source and target domains are aligned by the followed multi-branch network. After that, we perform two parts of information fusion: (1) We train a detection network shared by all branches based on the transferability of each source sample feature. The transferability of a source sample feature means the indistinguishable degree to the target domain sample features. (2) For using our model, the target sample features output by the multi-branch network are fused based on the average transferability of each domain. Moreover, we leverage both image-level and instance-level attention to promote positive cross-domain transfer and suppress negative transfer. Our main contributions are the two-stage feature alignments and information fusion. Extensive experimental results on various transfer scenarios show that our method achieves the state-of-the-art performance.     

多源域分布下优化权重的无监督迁移学习Boosting方法

深度决策树迁移学习Boosting方法（DTrBoost）可以有效地实现单源域有监督情况下向一个目标域迁移学习,但无法实现多个源域情况下的无监督迁移场景。针对这一问题,提出了多源域分布下优化权重的无监督迁移学习Boosting方法,主要思想是根据不同源域与目标域分布情况计算出对应的KL值,通过比较选择合适数量的不同源域样本训练分类器并对目标域样本打上伪标签。最后,依照各个不同源域的KL距离分配不同的学习权重,将带标签的各个源域样本与带伪标签的目标域进行集成训练得到最终结果。对比实验表明,提出的算法实现了更好的分类精度并对不同的数据集实现了自适应效果,分类错误率平均下降2.4%,在效果最好的marketing数据集上下降6%以上。     

Generalized Transfer Subspace Learning Through Low-Rank Constraint

It is expensive to obtain labeled real-world visual data for use in training of supervised algorithms. Therefore, it is valuable to leverage existing databases of labeled data. However, the data in the source databases is often obtained under conditions that differ from those in the new task. Transfer learning provides techniques for transferring learned knowledge from a source domain to a target domain by finding a mapping between them. In this paper, we discuss a method for projecting both source and target data to a generalized subspace where each target sample can be represented by some combination of source samples. By employing a low-rank constraint during this transfer, the structure of source and target domains are preserved. This approach has three benefits. First, good alignment between the domains is ensured through the use of only relevant data in some subspace of the source domain in reconstructing the data in the target domain. Second, the discriminative power of the source domain is naturally passed on to the target domain. Third, noisy information will be filtered out during knowledge transfer. Extensive experiments on synthetic data, and important computer vision problems such as face recognition application and visual domain adaptation for object recognition demonstrate the superiority of the proposed approach over the existing, well-established methods.     

一种面向多源领域的实例迁移学习

在迁移学习最大的特点就是利用相关领域的知识来帮助完成目标领域中的学习任务,它能够有效地在相似的领域或任务之间进行信息的共享和迁移,使传统的从零开始的学习变成可积累的学习,具有成本低、效率高等优点.针对源领域数据和目标领域数据分布类似的情况,提出一种基于多源动态TrAdaBoost的实例迁移学习方法.该方法考虑多个源领域知识,使得目标任务的学习可以充分利用所有源领域信息,每次训练候选分类器时,所有源领域样本都参与学习,可以获得有利于目标任务学习的有用信息,从而避免负迁移的产生.理论分析验证了所提算法较单源迁移的优势,以及加入动态因子改善了源权重收敛导致的权重熵由源样本转移到目标样本的问题.实验结果验证了此算法在提高识别率方面的优势.     

Conditional distribution-guided adversarial transfer learning network with multi-source domains for rolling bearing fault diagnosis

The application of transfer learning to effectively identify rolling bearing fault has been attracting much attention. Most of the current studies are based on single-source domain or multi-source domains constructed from different working conditions of the same machine. However, in practical scenarios, it is common to obtain multiple source domains from different machines, which brings new challenges to how to use these source domains to complete fault diagnosis. To solve the issue, a conditional distribution-guided adversarial transfer learning network with multi-source domains (CDGATLN) is developed for fault diagnosis of bearing installed on different machines. Firstly, the knowledge of multi-source domains from different machines is transferred to the single target domain by decreasing data distribution discrepancy between each source domain and target domain. Then, a conditional distribution-guided alignment strategy is introduced to decrease conditional distribution discrepancy and calculate the importance per source domain based on the conditional distribution discrepancy, so as to promote the knowledge transfer of each source domain. Finally, a monotone importance specification mechanism is constructed to constrain each importance to ensure that the source domain with low importance will not be discarded, which enables the knowledge of each source domain to participate in the construction of the model. Extensive experimental results verify the effectiveness and superiority of CDGATLN.     

一种基于局部分类精度的多源在线迁移学习算法

近年来,迁移学习得到越来越多的关注.现有的在线迁移学习算法一般从单个源领域迁移知识,然而,当源领域与目标领域相似度较低时,很难进行有效的迁移学习.基于此,提出了一种基于局部分类精度的多源在线迁移学习方法——LC-MSOTL.LC-MSOTL存储多个源领域分类器,计算新到样本与目标领域已有样本之间的距离以及各源领域分类器对其最近邻样本的分类精度,从源领域分类器中挑选局部精度最高的分类器与目标领域分类器加权组合,从而实现多个源领域知识到目标领域的迁移学习.在人工数据集和实际数据集上的实验结果表明,LC-MSOTL能够有效地从多个源领域实现选择性迁移,相对于单源在线迁移学习算法OTL,显示出了更高的分类准确率.     

一种异构直推式迁移学习算法

目标领域已有类别标注的数据较少时会影响学习性能,而与之相关的其他源领域中存在一些已标注数据.迁移学习针对这一情况,提出将与目标领域不同但相关的源领域上学习到的知识应用到目标领域.在实际应用中,例如文本-图像、跨语言迁移学习等,源领域和目标领域的特征空间是不相同的,这就是异构迁移学习.关注的重点是利用源领域中已标注的数据来提高目标领域中未标注数据的学习性能,这种情况是异构直推式迁移学习.因为源领域和目标领域的特征空间不同,异构迁移学习的一个关键问题是学习从源领域到目标领域的映射函数.提出采用无监督匹配源领域和目标领域的特征空间的方法来学习映射函数.学到的映射函数可以把源领域中的数据在目标领域中重新表示.这样,重表示之后的已标注源领域数据可以被迁移到目标领域中.因此,可以采用标准的机器学习方法(例如支持向量机方法)来训练分类器,以对目标领域中未标注的数据进行类别预测.给出一个概率解释以说明其对数据中的一些噪声是具有鲁棒性的.同时还推导了一个样本复杂度的边界,也就是寻找映射函数时需要的样本数.在4个实际的数据库上的实验结果,展示了该方法的有效性.     

Class conditional distribution alignment for domain adaptation

In this paper, we study the problem of domain adaptation, which is a crucial ingredient in transfer learning with two domains, that is, the source domain with labeled data and the target domain with none or few labels. Domain adaptation aims to extract knowledge from the source domain to improve the performance of the learning task in the target domain. A popular approach to handle this problem is via adversarial training, which is explained by the $\mathcal H \Delta \mathcal H$-distance theory. However, traditional adversarial network architectures just align the marginal feature distribution in the feature space. The alignment of class condition distribution is not guaranteed. Therefore, we proposed a novel method based on pseudo labels and the cluster assumption to avoid the incorrect class alignment in the feature space. The experiments demonstrate that our framework improves the accuracy on typical transfer learning tasks.     

Visual domain adaptation via transfer feature learning

One of the serious challenges in computer vision and image classification is learning an accurate classifier for a new unlabeled image dataset, considering that there is no available labeled training data. Transfer learning and domain adaptation are two outstanding solutions that tackle this challenge by employing available datasets, even with significant difference in distribution and properties, and transfer the knowledge from a related domain to the target domain. The main difference between these two solutions is their primary assumption about change in marginal and conditional distributions where transfer learning emphasizes on problems with same marginal distribution and different conditional distribution, and domain adaptation deals with opposite conditions. Most prior works have exploited these two learning strategies separately for domain shift problem where training and test sets are drawn from different distributions. In this paper, we exploit joint transfer learning and domain adaptation to cope with domain shift problem in which the distribution difference is significantly large, particularly vision datasets. We therefore put forward a novel transfer learning and domain adaptation approach, referred to as visual domain adaptation (VDA). Specifically, VDA reduces the joint marginal and conditional distributions across domains in an unsupervised manner where no label is available in test set. Moreover, VDA constructs condensed domain invariant clusters in the embedding representation to separate various classes alongside the domain transfer. In this work, we employ pseudo target labels refinement to iteratively converge to final solution. Employing an iterative procedure along with a novel optimization problem creates a robust and effective representation for adaptation across domains. Extensive experiments on 16 real vision datasets with different difficulties verify that VDA can significantly outperform state-of-the-art methods in image classification problem.     

Robust non-negative matrix factorization via joint sparse and graph regularization for transfer learning

In real-world applications, we often have to deal with some high-dimensional, sparse, noisy, and non-independent identically distributed data. In this paper, we aim to handle this kind of complex data in a transfer learning framework, and propose a robust non-negative matrix factorization via joint sparse and graph regularization model for transfer learning. First, we employ robust non-negative matrix factorization via sparse regularization model (RSNMF) to handle source domain data and then learn a meaningful matrix, which contains much common information between source domain and target domain data. Second, we treat this learned matrix as a bridge and transfer it to target domain. Target domain data are reconstructed by our robust non-negative matrix factorization via joint sparse and graph regularization model (RSGNMF). Third, we employ feature selection technique on new sparse represented target data. Fourth, we provide novel efficient iterative algorithms for RSNMF model and RSGNMF model and also give rigorous convergence and correctness analysis separately. Finally, experimental results on both text and image data sets demonstrate that our REGTL model outperforms existing start-of-art methods.     

基于多对多生成对抗网络的非对称跨域迁移行人再识别

无监督跨域迁移学习是行人再识别中一个非常重要的任务. 给定一个有标注的源域和一个没有标注的目标域, 无监督跨域迁移的关键点在于尽可能地把源域的知识迁移到目标域. 然而, 目前的跨域迁移方法忽略了域内各视角分布的差异性, 导致迁移效果不好. 针对这个缺陷, 本文提出了一个基于多视角的非对称跨域迁移学习的新问题. 为了实现这种非对称跨域迁移, 提出了一种基于多对多生成对抗网络(Many-to-many generative adversarial network, M2M-GAN)的迁移方法. 该方法嵌入了指定的源域视角标记和目标域视角标记作为引导信息, 并增加了视角分类器用于鉴别不同的视角分布, 从而使模型能自动针对不同的源域视角和目标域视角组合采取不同的迁移方式. 在行人再识别基准数据集Market1501、DukeMTMC-reID和MSMT17上, 实验验证了本文的方法能有效提升迁移效果, 达到更高的无监督跨域行人再识别准确率.     

多核局部领域适应学习

领域适应(或跨领域)学习旨在利用源领域(或辅助领域)中带标签样本来学习一种鲁棒的目标分类器,其关键问题在于如何最大化地减小领域间的分布差异.为了有效解决领域间特征分布的变化问题,提出一种三段式多核局部领域适应学习(multiple kernel local leaning-based domain adaptation,简称MKLDA)方法:1)基于最大均值差(maximum mean discrepancy,简称MMD)度量准则和结构风险最小化模型,同时,学习一个再生多核Hilbert空间和一个初始的支持向量机(support vector machine,简称SVM),对目标领域数据进行初始划分;2)在习得的多核Hilbert空间,对目标领域数据的类别信息进行局部重构学习;3)最后,利用学习获得的类别信息,在目标领域训练学习一个鲁棒的目标分类器.实验结果显示,所提方法具有优化或可比较的领域适应学习性能.