AI Paper Atlas

Research map / Label-efficient and robust learning / Domain adaptation and generalization

KD3A: Unsupervised Multi-Source Decentralized Domain Adaptation via Knowledge Distillation

Haozhe Feng, Zhaoyang You, Minghao Chen, Tianye Zhang, Minfeng Zhu, Fei Wu, Chao Wu, Wei Chen

ICML 2021 · Label-efficient and robust learning › Domain adaptation and generalization · approach: Source-free domain adaptation · 46 citations (OpenAlex)

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 changedWhereMeasuredEffect
Use KD3A instead of the strongest listed baselinemulti-source decentralized domain adaptation · DomainNet averageaccuracy (%): 48.7 -> 51.1Improved
Add Knowledge Vote to the full method (ablation)KD3A · UMDA target domainsaccuracyImproved
Add Consensus Focus to the full method (ablation)KD3A · UMDA target domainsaccuracyImproved
Add BatchNorm MMD to the full method (ablation)KD3A · UMDA target domainsaccuracyImproved
Use mixup data augmentation (ablation)KD3A · DomainNet averageaccuracy (%): 48.8 -> 51.1Improved
Use Consensus Focus instead of H-divergence weighting (ablation)KD3A source weighting · DomainNet average, IR-qdraccuracy (%): 57.9 -> 58.1Mixed
Use Consensus Focus instead of Info Gain (ablation)KD3A source weighting · DomainNet average, IR-qdraccuracy (%): 57.7 -> 58.1Improved
Use Consensus Focus instead of H-divergence weighting (ablation)KD3A source weighting · DomainNet average, MA-30accuracy (%): 49.8 -> 51.1Improved
Use Consensus Focus instead of Info Gain (ablation)KD3A source weighting · DomainNet average, MA-30accuracy (%): 48.9 -> 51.1Improved
Use Consensus Focus instead of dropping the bad source (ablation)KD3A source weighting · DomainNet average, MA-30accuracy (%): 50.7 -> 51.1Improved
Use Consensus Focus instead of dropping the irrelevant source (ablation)KD3A source weighting · DomainNet average, IR-qdraccuracy (%): 58.3 -> 58.1Hurt
Reduce communication rounds from r=100 to r=1 (ablation)KD3A · DomainNet averageaccuracy (%): 52.0 -> 51.1Hurt

Later papers that compared with this method

2 later accepted papers report KD3A as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy