The goal of unsupervised domain adaptation aims to utilize labeled data from source domain to annotate the target-domain data, which has none of the labels. Existing work uses Siamese network-based models to minimize the domain discrepancy to learn a domain-invariant feature. Alignment of the second-order statistics (covariances) of source and target distributions has been proven an effective method. Previous papers use Euclidean methods or geodesic methods (log-Euclidean) to measure the distance. However, covariances lay on a Riemannian manifold, and both methods cannot accurately calculate the Riemannian distance, so they cannot align the distribution well. To tackle the distribution alignment problem, this paper proposes mapped correlation alignment (MCA), a novel technique for end-to-end domain adaptation with deep neural networks. This method maps covariances from Riemannian manifold to reproducing kernel Hilbert space and uses Gaussian radial basis function-based positive definite kernels on manifolds to calculate the inner product on reproducing kernel Hilbert space, and then uses Euclidean metric accurate measuring the distance to align the distribution better. This paper builds an end-to-end model to minimize both the classification loss and the MCA loss. The model can be trained efficiently using backpropagation. Experiments show that the MCA method yields the state-of-the-art results on standard domain adaptation data sets.

The training of deep neural networks relies on massive high-quality labeled data which is expensive in practice. To tackle this problem, domain adaptation is proposed to transfer knowledge from label-rich source domain to unlabeled target domain to learn a classifier that can well classify target data. However, people don't consider privacy issues in domain adaptation. In this paper, we introduce a novel method that builds an effective model without sharing sensitive data between source and target domain. Target domain party can benefit from label-rich source domain without revealing its privacy data. We transfer the traditional domain adaptation into a federated setting, where a global server contains a shared global model. Additionally, homomorphic encryption (HE) algorithm is used to guarantee the computing security. Experiments show that our method performs effectively without reducing the accuracy. Our method can achieve secure knowledge transfer and privacy-preserving domain adaptation.

Cross-domain underwater object recognition is essential for intelligent visual monitoring in marine ranching, yet domain shift caused by varying water conditions and imaging devices severely degrades model performance. Existing adversarial domain adaptation methods align feature distributions to mitigate domain shift, but they often fail to preserve the fine-grained discriminative structure required for distinguishing visually similar marine species. Discriminator-free adversarial domain adaptation constrains only the sum of singular values, leading to projection distortion, thereby degrading target discriminability. We observe that the discriminative structure of the source domain is encoded in the singular value distribution of classifier outputs, and aligning this distribution across domains, rather than its sum, preserves discriminability during knowledge transfer. Based on this insight, we propose Adversarial Spectral Distribution Alignment (ASDA). ASDA consists of Spectral Distribution Alignment (SDA), which minimizes the Wasserstein distance between source and target singular value distributions, and a Dynamic Feature Queue (DFQ) with an adaptive length schedule that provides stable spectral distribution estimates across mini-batches. By enforcing singular value ratio consistency across all principal directions, SDA achieves fine-grained alignment, which reduces domain discrepancy while preserving discriminative features. Experimental results on two underwater image datasets demonstrate that ASDA outperforms existing domain adaptation methods.