Abstract:Adversarial learning has proven to be an effective method for capturing transferable features for unsupervised domain adaptation. However, some existing conditional adversarial domain adaptation methods assign equal importance to different samples, ignoring the fact that hard-to-transfer samples might damage the conditional adversarial adaptation procedure. Meanwhile, some methods can only roughly align marginal distributions across domains, but cannot ensure category distributions alignment, causing classifiers to make uncertain or even wrong predictions for some target data. Furthermore, we find that the feature norms of real images usually follow a complex distribution, so directly matching the mean feature norms of two domains cannot effectively reduce the statistical discrepancy of feature norms and may potentially induce feature degradation. In this paper, we develop a Trust-aware Conditional Adversarial Domain Adaptation (TCADA) method for solving the aforementioned issues. To quantify data transferability, we suggest utilizing posterior probability modeled by a Gaussian-uniform mixture, which effectively facilitates conditional domain alignment. Based on this posterior probability, a confidence-guided alignment strategy is presented to promote precise alignment of category distributions and accelerate the learning of shared features. Moreover, a novel optimal transport-based strategy is introduced to align the feature norms and facilitate shared features becoming more informative. To encourage classifiers to make more accurate predictions for target data, we also design a mixed information-guided entropy regularization term to promote deep features being away from the decision boundaries. Extensive experiments show that our method greatly improves transfer performance on various tasks.

Transfer Kernel Sparse Coding Based on Dynamic Distribution Alignment for Image Representation

Trust-aware Conditional Adversarial Domain Adaptation with Feature Norm Alignment.

Joint Domain Alignment and Discriminative Feature Learning for Unsupervised Deep Domain Adaptation

Transfer Sparse Coding for Robust Image Representation

LSDT: Latent Sparse Domain Transfer Learning for Visual Adaptation

Domain Invariant Transfer Kernel Learning.

Domain Adaptation Based on Semi-Supervised Cross-Domain Mean Discriminative Analysis and Kernel Transfer Extreme Learning Machine

Discriminant Kernel Assignment for Image Coding

Dictionary Transfer for Image Denoising Via Domain Adaptation

Cross-Domain Sparse Coding

Kernel optimization-based discriminant analysis for face recognition

Subspace Metric-Based Transfer Learning for Spindle Thermal Error Prediction under Time-Varying Conditions

A novel kernel collaborative representation approach for image classification

Equiangular Kernel Dictionary Learning with Applications to Dynamic Texture Analysis

Cross-Domain Kernel Induction for Transfer Learning

Domain Adaptation via Transfer Component Analysis

Iterative joint classifier and domain adaptation for visual transfer learning

Dynamic Kernel-Based Adaptive Spatial Aggregation for Learned Image Compression

Transfer Learning with Dynamic Distribution Adaptation

Deep Sparse Informative Transfer SoftMax for Cross-Domain Image Classification.