Research map / Federated learning and privacy / Personalized and heterogeneous FL
Federated Learning with Only Positive Labels
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 server-side spreadout regularization with stochastic negative mining | positive-only federated training · CIFAR-10 and CIFAR-100 | Precision@1 (%): 10.7 -> 86.3; 9.8 -> 92.4; 1.0 -> 67.9; 1.1 -> 69.6 | Helped |
| add server-side spreadout regularization with stochastic negative mining | positive-only federated training · AmazonCat | P@1 (%): 3.4 -> 92.1 | Helped |
| add server-side spreadout regularization with stochastic negative mining | positive-only federated training · Amazon670K | P@1 (%): 0.0 -> 33.1 | Helped |
| add server-side spreadout regularization with stochastic negative mining | positive-only federated training · WikiLSHTC | P@1 (%): 7.6 -> 37.2 | Helped |
| fix class embeddings at random initialization instead of training them | positive-only federated training · CIFAR-10, ResNet-32 | Precision@1 (%): 92.1 vs 92.4 | No clear effect |
| fix class embeddings at random initialization instead of training them | positive-only federated training · Amazon670K and WikiLSHTC | P@1 (%): 4.3 vs 33.1; 7.9 vs 37.2 | Hurt |
| use k = 10 confusing labels (ablation) | FedAwS · AmazonCat | P@1 (%): 92.1 -> 26.3 | Hurt |
| use k = 500 confusing labels rather than k = 100 (ablation) | FedAwS · AmazonCat | P@1 (%): 92.1 -> 86.9 | Hurt |
| use all labels rather than stochastic negative mining (ablation) | FedAwS · AmazonCat | P@1 (%): 92.1 -> 87.7 | Hurt |
| set the spreadout learning-rate multiplier to λ = 1 instead of λ = 10 (ablation) | FedAwS · AmazonCat | P@1 (%): 92.1 -> 73.2 | Hurt |
| set the spreadout learning-rate multiplier to λ = 100 instead of λ = 10 (ablation) | FedAwS · AmazonCat | P@1 (%): 92.1 -> 92.2 | No clear effect |
| normalize class and instance embeddings to unit L2 norm (ablation) | FedAwS · the reported experiments | model quality | Helped |
Closest papers by meaning
- (FL)$^2$: Overcoming Few Labels in Federated Semi-Supervised Learning (NeurIPS 2024 · similarity 0.86)
- Exploring Vacant Classes in Label-Skewed Federated Learning (AAAI 2025 · similarity 0.86)
- FedUV: Uniformity and Variance for Heterogeneous Federated Learning (CVPR 2024 · similarity 0.86)
- Progressive Distribution Matching for Federated Semi-Supervised Learning (AAAI 2025 · similarity 0.85)
- Exploiting Label Skews in Federated Learning with Model Concatenation (AAAI 2024 · similarity 0.84)
- Combating Data Imbalances in Federated Semi-supervised Learning with Dual Regulators (AAAI 2024 · similarity 0.84)
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.