Research map / Label-efficient and robust learning / Domain adaptation and generalization
Information-Theoretic Regularization for Multi-Source Domain Adaptation
Geon Yeong Park, Sang Wan Lee
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace M3SDA-beta with MIAN (single unified domain discriminator) | multi-source DA · Digits-Five avg | accuracy: 87.65 -> 92.03 | Improved |
| replace M3SDA-beta with MIAN | multi-source DA · Digits-Five MNIST-M | accuracy: 72.82 -> 84.36 | Improved |
| replace M3SDA-beta with MIAN | multi-source DA · Office-31 avg | accuracy: 89.45 -> 90.87 | Improved |
| replace M3SDA with MIAN | multi-source DA · Office-Home avg | accuracy: 71.39 -> 73.12 | Improved |
| add decaying batch spectral penalization (MIAN-gamma) to MIAN | MIAN · Office-31 avg | accuracy: 90.87 -> 91.26 | Improved |
| add decaying batch spectral penalization (MIAN-gamma) to MIAN | MIAN · Office-Home avg | accuracy: 73.12 -> 74.11 | Improved |
| use one unified domain discriminator instead of multiple discriminators (Multi D) (ablation) | MIAN · MNIST-M and SVHN targets | accuracy | Improved |
| use one unified domain discriminator instead of multiple discriminators (ablation) | MIAN · MNIST-M and SVHN targets | gradient variance | Improved |
| use binary cross entropy instead of multi-class cross entropy for the discriminator objective (ablation) | MIAN · Digits-Five | accuracy | Improved |
| use binary cross entropy instead of multi-class cross entropy for the discriminator objective (ablation) | MIAN · Digits-Five | proxy A-distance / mutual information | Improved |
| add decaying spectral penalization and annealed information regularization (ablation) | MIAN · Office-31 (DSLR to Amazon) | accuracy | Improved |
| use constant spectral penalization (BSP) instead of decaying (ablation) | MIAN · Office-31 low-shift targets (Webcam, DSLR) | accuracy | Hurt |
Closest papers by meaning
- Adversarial Multiple Source Domain Adaptation (NeurIPS 2018 · similarity 0.91)
- Multi-Adversarial Domain Adaptation (AAAI 2018 · similarity 0.90)
- Multi-Source Distilling Domain Adaptation (AAAI 2020 · similarity 0.89)
- Deep Cocktail Network: Multi-Source Unsupervised Domain Adaptation With Category Shift (CVPR 2018 · similarity 0.89)
- f-Domain Adversarial Learning: Theory and Algorithms (ICML 2021 · similarity 0.89)
- A General Representation-Based Approach to Multi-Source Domain Adaptation (ICML 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
