Heterogeneous Domain Adaptation via Soft Transfer Network

Abstract

Heterogeneous domain adaptation (HDA) aims to facilitate the learning task in a target domain by borrowing knowledge from a heterogeneous source domain. In this paper, we propose a Soft Transfer Network (STN), which jointly learns a domain-shared classifier and a domain-invariant subspace in an end-to-end manner, for addressing the HDA problem. The proposed STN not only aligns the discriminative directions of domains but also matches both the marginal and conditional distributions across domains. To circumvent negative transfer, STN aligns the conditional distributions by using the soft-label strategy of unlabeled target data, which prevents the hard assignment of each unlabeled target data to only one category that may be incorrect. Further, STN introduces an adaptive coefficient to gradually increase the importance of the soft-labels since they will become more and more accurate as the number of iterations increases. We perform experiments on the transfer tasks of image-to-image, text-to-image, and text-to-text. Experimental results testify that the STN significantly outperforms several state-of-the-art approaches.

CHEER: Rich Model Helps Poor Model via Knowledge Infusion

“…Domain Adaptation: There also exists another body of non-deep learning transfer paradigms that were often referred to as domain adaption. This however often include methods that not only assume access domain-specific [24,29,[36][37][38] and/or model-specific knowledge of the domains being adapted [17,20,25,27,28,35,41,43], but are also not applicable to deep learning models [10,39] with arbitrary architecture as addressed in our work.…”

Prototype-Matching Graph Network for Heterogeneous Domain Adaptation

“…It has been witnessing an explosion of multimedia data on the web in recent decades [5,23,32,35,40]. Unfortunately, these data are hard to be fully exploited without an expensive and tedious human annotation process.…”

“…Most of the existing DA methods are designed for the homogeneous scenario [2,3,10,21], where the cross-domain features are of the same type. More recently, Heterogeneous Domain Adaptation (HDA) is proposed, where the cross-domain features can be represented by different types of features from various modalities [19,20,38,40,43]. Existing HDA methods can be roughly categorized as supervised [9,14,17,31] or semi-supervised methods [4,18,19,40,42].…”

“…More recently, Heterogeneous Domain Adaptation (HDA) is proposed, where the cross-domain features can be represented by different types of features from various modalities [19,20,38,40,43]. Existing HDA methods can be roughly categorized as supervised [9,14,17,31] or semi-supervised methods [4,18,19,40,42]. Typically, the supervised methods exploit the labeled source domain and a limited number of labeled target domain samples to train the learning algorithm.…”

“…Despite the great progress that has been achieved in the area of HDA, the existing approaches are limited in a sense that they do not explicitly maximize the margin among different classes of the target domain, so that the decision boundaries are not adequately discriminative. In particular, [19] aims to train an adaptable classifier without minimizing the domain discrepancy; [39] reduces the domain shift by aligning the cross-domain manifolds; [30,34,40] apply the Maximum Mean Discrepancy (MMD) [11] and its variants to mitigate the domain discrepancy and conditional distribution discrepancy.…”