Research map / Label-efficient and robust learning / Domain adaptation and generalization
Deep Cocktail Network: Multi-Source Unsupervised Domain Adaptation With Category Shift
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Add online hard domain batch mining (ablation) | DCTN · Office-31 A,W→D, vanilla | accuracy (%): 99.0 -> 99.6 | Helped |
| Add online hard domain batch mining (ablation) | DCTN · Office-31 A,D→W, vanilla | accuracy (%): 96.1 -> 96.9 | Helped |
| Add online hard domain batch mining (ablation) | DCTN · Office-31 D,W→A, vanilla | accuracy (%): 55.0 -> 54.9 | No clear effect |
| Add online hard domain batch mining (ablation) | DCTN · Office-31 A,D→W, category shift overlap | accuracy (%): 89.3 -> 90.2 | Helped |
| Add online hard domain batch mining (ablation) | DCTN · Office-31 A,D→W, category shift disjoint | accuracy (%): 82.6 -> 82.9 | Helped |
| Remove pseudo-labeling and train the category classifier on source labels only (ablation) | DCTN adversarial-only variant · Office-31 A,D→W | accuracy | Hurt |
| Remove the adversary and use average multi-source predictions (ablation) | DCTN pseudo-only variant · Office-31 A,D→W | accuracy | Hurt |
| Use multi-source adaptation with DCTN rather than the source-only model | Office-31 A,D→W, category shift overlap · Office-31 A,D→W, category shift overlap | accuracy: 84.4 -> 90.2 | Helped |
| Use multi-source adaptation with DCTN rather than the source-only model | Office-31 A,D→W, category shift disjoint · Office-31 A,D→W, category shift disjoint | accuracy: 78.1 -> 82.9 | Helped |
| Use multi-source adaptation with DCTN rather than the source-only model | ImageCLEF-DA I,P→C, category shift overlap · ImageCLEF-DA I,P→C, category shift overlap | accuracy: 86.3 -> 88.7 | Helped |
| Use multi-source adaptation with DCTN rather than the source-only model | ImageCLEF-DA I,P→C, category shift disjoint · ImageCLEF-DA I,P→C, category shift disjoint | accuracy: 81.5 -> 82.0 | Helped |
| Apply DCTN instead of RevGrad source-combine adaptation | Digits-five · Digits-five mm, mt, sy, up→sv | accuracy (%): 68.9 -> 77.5 | Helped |
Closest papers by meaning
- Multi-Source Distilling Domain Adaptation (AAAI 2020 · similarity 0.92)
- Universal Source-Free Domain Adaptation (CVPR 2020 · similarity 0.90)
- Blending-Target Domain Adaptation by Adversarial Meta-Adaptation Networks (CVPR 2019 · similarity 0.89)
- Adversarial Multiple Source Domain Adaptation (NeurIPS 2018 · similarity 0.89)
- On Balancing Bias and Variance in Unsupervised Multi-Source-Free Domain Adaptation (ICML 2023 · similarity 0.89)
- Can We Evaluate Domain Adaptation Models Without Target-Domain Labels? (ICLR 2024 · 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.