Graph Adaptive Knowledge Transfer for Unsupervised Domain Adaptation

Abstract

Unsupervised domain adaptation has caught appealing attentions as it facilitates the unlabeled target learning by borrowing existing well-established source domain knowledge. Recent practice on domain adaptation manages to extract effective features by incorporating the pseudo labels for the target domain to better solve cross-domain distribution divergences. However, existing approaches separate target label optimization and domain-invariant feature learning as different steps. To address that issue, we develop a novel Graph Adaptive Knowledge Transfer (GAKT) model to jointly optimize target labels and domain-free features in a unified framework. Specifically, semi-supervised knowledge adaptation and label propagation on target data are coupled to benefit each other, and hence the marginal and conditional disparities across different domains will be better alleviated. Experimental evaluation on two cross-domain visual datasets demonstrates the effectiveness of our designed approach on facilitating the unlabeled target task learning, compared to the state-of-the-art domain adaptation approaches.

Learning Target-Domain-Specific Classifier for Partial Domain Adaptation

“…Sun et al [25] proposed to minimize domain bias by matching the mean and covariance of the distributions simultaneously. Some other works also explore weighted strategies in domain alignment [31], [32]. Ding et al [31] proposed an iterative refinement scheme to optimize the probabilistic and class-wise subspace adaptation term from a graph-based label propagation perspective.…”

“…Some other works also explore weighted strategies in domain alignment [31], [32]. Ding et al [31] proposed an iterative refinement scheme to optimize the probabilistic and class-wise subspace adaptation term from a graph-based label propagation perspective. In [32], two coupled deep neural networks were used to extract the representative features, then a weighted class-wise adaptation scheme was built for minimizing the distribution gap.…”

d-SNE: Domain Adaptation Using Stochastic Neighborhood Embedding

“…Domain multi-task learning: In order to improve the discriminative capabilities of feature representations, Tzeng et al introduced a shared feature extractor for both source and target domain with three different losses in a multitask learning manner [27]. Ding et al uses a knowledge graph model to jointly optimize target labels with domainfree features in a unified framework [4]. These losses also acted as a strong regularizer.…”

AdaGraph: Unifying Predictive and Continuous Domain Adaptation Through Graphs

“…Finally, our work is linked to graph-based domain adaptation methods [6,5]. Differently from these works however, in our approach a node does not represent a single sample but a whole domain and edges do not link semantically related samples but domains with related metadata.…”