Research map / Federated learning and privacy / Communication-efficient FL
Outsourcing Training without Uploading Data via Efficient Collaborative Open-Source Sampling
Junyuan Hong, Lingjuan Lyu, Jiayu Zhou, Michael Spranger
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 |
|---|---|---|---|
| add OoD samples to cloud data (ID -> ID+OoD) | random sampling budget 1000 · Digits average | accuracy (%): 70.4 -> 62.4 | Hurt |
| increase sampling budget with ID+OoD cloud data | random sampling · Digits average | accuracy (%): 62.4 -> 68.4 | Improved |
| use ECOS instead of random sampling in selective labeling | ECOS · DomainNet Quickdraw, budget 1000 | accuracy (%): 77.9 -> 84.3 | Improved |
| use K-Center instead of random sampling | K-Center · DomainNet Quickdraw, budget 3000 | accuracy (%): 86.1 -> 60.5 | Hurt |
| use ECOS instead of K-Center in selective labeling | ECOS · Digits SVHN, budget 2000 | accuracy (%): 65.1 -> 68.7 | Improved |
| use ECOS in adaptive model compression | ECOS · DomainNet compression, budget 1000 | avg accuracy (%): 76.4 -> 78.6 | Improved |
| use ECOS before private kNN client labeling | ECOS · Digits client labeling, budget 5000 | avg accuracy (%): 65.9 -> 69.0 | Improved |
| use ECOS before private kNN client labeling | ECOS · DomainNet client labeling, budget 3000 | avg accuracy (%): 40.4 -> 43.2 | Improved |
| add proximity objective only (no diversity) (ablation) | ECOS · DomainNet client labeling | avg accuracy (%): 40.4 -> 35.2 | Hurt |
| add diversity objective only (no proximity) (ablation) | ECOS · DomainNet client labeling | avg accuracy (%): 40.4 -> 42.0 | Improved |
| combine proximity and diversity (ablation) | ECOS · DomainNet client labeling | avg accuracy (%): 42.0 -> 43.2 | Improved |
| increase compression size R at large budget (e.g. 8000) (ablation) | ECOS · Digits ID+OoD | ID TPR | Mixed |
Closest papers by meaning
- Small Scale Data-Free Knowledge Distillation (CVPR 2024 · similarity 0.82)
- Outsourced Diffusion Sampling: Efficient Posterior Inference in Latent Spaces of Generative Models (ICML 2025 · similarity 0.82)
- GraSS: Scalable Data Attribution with Gradient Sparsification and Sparse Projection (NeurIPS 2025 · similarity 0.82)
- DeepSpeed Data Efficiency: Improving Deep Learning Model Quality and Training Efficiency via Efficient Data Sampling and Routing (AAAI 2024 · similarity 0.82)
- Unextractable Protocol Models: Collaborative Training and Inference without Weight Materialization (NeurIPS 2025 · similarity 0.82)
- DOS: Diverse Outlier Sampling for Out-of-Distribution Detection (ICLR 2024 · similarity 0.82)
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.
