Posts by Collection
portfolio
publications
Letting Go of Self-Domain Awareness: Multi-Source Domain-Adversarial Generalization via Dynamic Domain-Weighted Contrastive Transfer Learning
Published in ECAI, 2023
Domain generalization (DG) aims to train models that generalize to unseen target domains, often by learning domain-invariant representations. However, overly compressed representations can confuse classes within the same domain. To address this, we propose MsCtrl, a framework incorporating dynamic domain-weighted...
Recommended citation: Y. Ma et al., " Letting Go of Self-Domain Awareness: Multi-Source Domain-Adversarial Generalization via Dynamic Domain-Weighted Contrastive Transfer Learning," in ECAI 2023. IOS Press, 2023: 1664-1671.
Download Paper
Class-imbalanced Time Series Adaptation via Multi-Expert Consistency Entropy Minimization
Published in IEEE International Conference on Bioinformatics and Biomedicine, 2024
Wearable-based time series recognition aims to infer behavioral classes based on time series signals collected by sensors. Domain Adaptation (DA) methods have significantly improved the performance on out-of-domain test classification tasks. However, highly imbalanced label distribution in real-world tasks hinders...
Recommended citation: Y. Ma et al., "Class-imbalanced Time Series Adaptation via Multi-Expert Consistency Entropy Minimization," 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), Lisbon, Portugal, 2024, pp. 2284-2290, doi: 10.1109/BIBM62325.2024.10822753.
Download Paper
Grade-Skewed Domain Adaptation via Asymmetric Bi-Classifier Discrepancy Minimization for Diabetic Retinopathy Grading
Published in IEEE Transactions on Medical Imaging (TMI), 2025
Diabetic retinopathy (DR) is a leading cause of preventable blindness, with deep learning showing promise in its grading. However, domain shifts, small lesions, and imbalanced grade distributions complicate generalization and adaptation, often leading to biased predictions. To address this, we...
Recommended citation: Y. Ma et al., "Grade-Skewed Domain Adaptation via Asymmetric Bi-Classifier Discrepancy Minimization for Diabetic Retinopathy Grading," in IEEE Transactions on Medical Imaging, doi: 10.1109/TMI.2024.3485064.
Download Paper
DDIR: Domain-Disentangled Invariant Representation Learning for Tailored Predictions
Published in Knowledge-Based Systems, 2025
Traditional training struggles with large datasets due to distributional differences. Domain generalization (DG) methods, like DIR learning, perform well with domain shifts but often hurt in-distribution performance. We propose DDIR learning, which preserves domain-orthogonal invariant (DOI) information without redundancy. DDIR...
Recommended citation: Y. Ma et al., "DDIR: Domain-Disentangled Invariant Representation Learning for Tailored Predictions," in Knowledge-Based Systems 2025.
Download Paper