Deep Cocktail Network: Multi-source Unsupervised Domain Adaptation with Category Shift

By Ruijia Xu, Ziliang Chen, Wangmeng Zuo, Junjie Yan, Liang Lin. 2018

437 citations recorded by the source corpus.

Unsupervised domain adaptation (UDA) conventionally assumes labeled source samples coming from a single underlying source distribution. Whereas in practical scenario, labeled data are typically collected from diverse sources. The multiple sources are different not only from the target but also from each other, thus, domain adaptater should not be modeled in the same way. Moreover, those sources may not completely share their categories, which further brings a new transfer challenge called category shift. In this paper, we propose a deep cocktail network (DCTN) to battle the domain and category shifts among multiple sources. Motivated by the theoretical results in [33], the target distribution can be represented as the weighted combination of source distributions, and, the multi-source UDA via DCTN is then performed as two alternating steps: i) It deploys multi-way adversarial learning to minimize the discrepancy between the target and each of the multiple source domains, which also obtains the source-specific perplexity scores to denote the possibilities that a target sample belongs to different source domains. ii) The multi-source category classifiers are integrated with the perplexity scores to classify target sample, and the pseudo-labeled target samples together with source samples are utilized to update the multi-source category classifier and the feature extractor. We evaluate DCTN in three domain adaptation benchmarks, which clearly demonstrate the superiority of our framework.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Radiology, Nuclear Medicine and Imaging

Citations in the corpus

Multiple-Source Adaptation with Domain Classifiers (2020, background): “More recently, Xu et al. [2018] deployed multi-way adversarial training to obtain a multi-way discriminator, and took a weighted sum of discriminator’s scores and domain-speciﬁc predictors to make predictions.”
Learngene: From Open-World to Your Learning Task (2021, background): “However, a network, trained in a speciﬁc task, is incapable of recognizing unknown/unseen classes due to domain shift (Ben-David et al. 2007) and category shift (Xu et al. 2018).”
Moment Matching for Multi-Source Domain Adaptation (2018, methodology): “More applied works [6, 45] use an adversarial discriminator to align the multi-source domains with the target domain.”
Best Sources Forward: Domain Generalization through Source-Specific Nets (2018, background): “Such research efforts go under the name of Domain Adaptation, for which there is a large literature of shallow and deep approaches [2, 1, 19, 14].”
Domain Generalization by Solving Jigsaw Puzzles (2019, background): “To get closer to real world conditions, recent work has started to focus on cases where the source data are drawn from multiple distributions [30, 48] and the target covers only a part of the source classes [5, 1].”
Learning to Optimize Domain Specific Normalization for Domain Generalization (2019, methodology): “Finally, we employ five datasets—MNIST, MNIST-M, USPS, SVHN and Synthetic Digits— for digit recognition and split training and testing subsets following [30].”
AdaGraph: Unifying Predictive and Continuous Domain Adaptation Through Graphs (2019, background): “This fact has motivated the study of multi-source DA algorithms [34, 23].”
Domain Agnostic Learning with Disentangled Representations (2019, background): “Originating from the theoretical analysis in (Ben-David et al., 2010; Mansour et al., 2009; Cram-mer et al., 2008), MSDA has been applied to many practical applications (Xu et al., 2018; Duan et al., 2012; Peng et al., 2018).”
Domain Generalization Using a Mixture of Multiple Latent Domains (2019, methodology): “Although these methods assume single-source and target domains, multi-source domain adaptation methods (Xu et al. 2018; Schoenauer-Sebag et al. 2019) utilize multiple source domains for domain adaptation to learn domain relations.”
Domain2Vec: Domain Embedding for Unsupervised Domain Adaptation (2020, methodology): “Inspired by Xu et al [39], we compare MSDA results with two other evaluation standards: (i) single best, reporting the single best-preforming source transfer result on the test set, and (ii) source combine, combining the source domains to a single domain and performing traditional single-source sing”
Learning to Balance Specificity and Invariance for In and Out of Domain Generalization (2020, methodology): “In multi-source domain adaptation, the source dataset consists of a mixture of multiple domains where domain alignment is performed using an adversarial interplay involving a k-way domain discriminator in [50], and multi-domain moment matching in [38].”
Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation (2020, result): “For mutli-source and multi-target DA, we compare with DCTN (Xu et al., 2018), MCD (Saito et al., (Long et al., 2018; Deng et al., 2019; Xu et al., 2019; Peng et al., 2019a).”
A Review of Single-Source Deep Unsupervised Visual Domain Adaptation (2020, background): “#sources single-source DA NS = 1 there is only one source domain [36, 23, 11, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99”
A Survey of Unsupervised Deep Domain Adaptation (2018, methodology): “[227] survey multi-source domain adaptation, and since then a number of other methods [28, 88, 95, 157, 184, 192, 256, 259, 281] have been developed for this case.”
Minimum Class Confusion for Versatile Domain Adaptation (2019, methodology): “MDAN [54] provides theoretical insights for MSDA, while Deep Cocktail Network [48] (DCTN) and MSDA [31] extend adversarial training and moment-matching to MSDA, respectively.”
Aligning Domain-Specific Distribution and Classifier for Cross-Domain Classification from Multiple Sources (2019, methodology): “In (Xu et al. 2018), they propose a target classification operator to combine the multiple source classifiers.”