Research map / Federated learning and privacy / Personalized and heterogeneous FL
Neural Collapse-Inspired Multi-Label Federated Learning under Label-Distribution Skew
Can Peng, Yuyuan Liu, YINGYU YANG, Pramit Saha, Qianye Yang, J. Noble
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 predefined ETF classifier alone (ablation) | baseline FL model · DermaMNIST (beta=0.1) | macro-F1: 40.69 -> 33.35 | Hurt |
| add LADM class-specific extraction on top of ETF classifier (ablation) | ETF classifier · DermaMNIST (beta=0.1) | macro-F1: 33.35 -> 47.95 | Improved |
| add L_Neg and L_Pos regularisation (ablation) | ETF + LADM · DermaMNIST (beta=0.1) | macro-F1: 47.95 -> 51.38 | Improved |
| add only L_Neg (without L_Pos) (ablation) | ETF + LADM · DermaMNIST (beta=0.1) | macro-F1: 47.95 -> 45.06 | Hurt |
| add only L_Pos (without L_Neg) (ablation) | ETF + LADM · DermaMNIST (beta=0.1) | macro-F1: 47.95 -> 49.84 | Improved |
| add ETF classifier alone (ablation) | baseline FL model · CIFAR-10 (beta=0.5) | macro-F1: 40.65 -> 15.29 | Hurt |
| add LADM on top of ETF classifier (ablation) | ETF classifier · CIFAR-10 (beta=0.5) | macro-F1: 15.29 -> 44.64 | Improved |
| add L_Neg and L_Pos regularisation (ablation) | ETF + LADM · CIFAR-10 (beta=0.5) | macro-F1: 44.64 -> 49.22 | Improved |
| use fixed ETF queries instead of learnable random queries in LADM (ablation) | LADM · DermaMNIST (beta=0.1) | macro-F1: 47.94 -> 47.95 | No clear effect |
| use fixed ETF queries instead of learnable ETF-initialised queries (ablation) | LADM · DermaMNIST (beta=0.1) | micro-F1: 53.37 -> 60.49 | Improved |
| use fixed ETF queries instead of learnable ETF-initialised queries (ablation) | LADM · CIFAR-10 (beta=0.5) | macro-AUC: 86.26 -> 85.84 | Mixed |
| use FedNCA-ML instead of second-best method | FedLGT / SphereFed · CIFAR-10 (beta=0.5) | macro-AUC: 83.63 -> 87.55 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SCAFFOLD — SCAFFOLD: Stochastic Controlled Averaging for Federated Learning (ICML 2020) · better on 1 of 1 shared result
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
- Exploring Vacant Classes in Label-Skewed Federated Learning (AAAI 2025 · similarity 0.90)
- Exploiting Label Skews in Federated Learning with Model Concatenation (AAAI 2024 · similarity 0.90)
- Language-Guided Transformer for Federated Multi-Label Classification (AAAI 2024 · similarity 0.89)
- FedLoGe: Joint Local and Generic Federated Learning under Long-tailed Data (ICLR 2024 · similarity 0.88)
- FedWon: Triumphing Multi-domain Federated Learning Without Normalization (ICLR 2024 · similarity 0.88)
- (FL)$^2$: Overcoming Few Labels in Federated Semi-Supervised Learning (NeurIPS 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.
