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

Abstract

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 unsupervised domain adaptation 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 pseudolabeled 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.

Moment Matching for Multi-Source Domain Adaptation

“…Mansour et al [28] claim that the target hypothesis can be represented by a weighted combination of source hypotheses. In the more applied works, Deep Cocktail Network (DCTN) [45] proposes a k-way domain discriminator and category classifier for digit classification and real-world object recognition. Hoffman et al [14] propose normalized solutions with theoretical guarantees for cross-entropy loss, aiming to provide a solution for the MSDA problem with very practical benefits.…”

“…Ben-David et al [1] pioneer this direction by introducing an H∆H-divergence between the weighted combination of source domains and target domain. More applied works [6,45] use an adversarial discriminator to align the multi-source domains with the target domain. However, these works focus only on aligning the source domains with the target, neglecting the domain shift between the source domains.…”

“…Secondly, we propose a novel approach called M 3 SDA to tackle MSDA task by aligning the source domains with the target domain, and aligning the source domains with each other simultaneously. We dispose multiple complex adversarial training procedures presented in [45], but di-rectly align the moments of their deep feature distributions, leading to a more robust and effective MSDA model. To our best knowledge, we are the first to empirically demonstrate that aligning the source domains is beneficial for MSDA tasks.…”

Blending-Target Domain Adaptation by Adversarial Meta-Adaptation Networks

“…Benchmarks. Digit-five [44] is composed of five domain sets drawn from mt (MNIST) [23], mm (MNIST-M) [10], sv(SVHN) [32], up (USPS) and sy (Synthetic Digits) [10], respectively. There are 25000 for training and 9000 for testing in mt, mm, sv, sy, while the entire USPS is chosen as a domain set up.…”

AdaGraph: Unifying Predictive and Continuous Domain Adaptation Through Graphs

“…In many applications multiple source domains may be available. This fact has motivated the study of multi-source DA algorithms [34,23]. In [34] an adversarial learning framework for multi-source DA is proposed, inspired by [10].…”