Research map / Federated learning and privacy / Personalized and heterogeneous FL
A Framework for Double-Blind Federated Adaptation of Foundation Models
Nurbek Tastan, Karthik Nandakumar
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 |
|---|---|---|---|
| replace linear probing with BlindFed (parallel adapter) | ViT-Base · SVHN (centralized, pooled) | balanced accuracy: 0.5938 -> 0.9232 | Improved |
| replace linear probing with BlindFed | ViT-Base · CIFAR-100 (federated K=5, Dirichlet 0.01) | balanced accuracy: 0.5317 -> 0.6620 | Improved |
| replace linear probing with BlindFed | ViT-Base · CIFAR-10 (federated K=50, Dirichlet 1.0) | balanced accuracy: 0.9007 -> 0.9287 | Improved |
| add stochastic block sampling (SBS) to BlindFed | BlindFed · CIFAR-10 (centralized, pooled) | balanced accuracy: 0.9428 -> 0.9443 | No clear effect |
| add SBS to BlindFed | BlindFed · CIFAR-100 (centralized, pooled) | balanced accuracy: 0.7930 -> 0.7869 | No clear effect |
| add SBS to BlindFed | BlindFed · CIFAR-10 (federated K=50, Dirichlet 1.0) | balanced accuracy: 0.9287 -> 0.9388 | Improved |
| add SBS to BlindFed | BlindFed · Fed-ISIC2019 (federated K=5, aux Fed-ISIC2019 center 0, InD) | balanced accuracy: 0.6679 -> 0.6831 | Improved |
| add SBS to BlindFed | BlindFed · CIFAR-10 (federated K=5, Dirichlet 0.01) | balanced accuracy: 0.8540 -> 0.8489 | Hurt |
| use out-of-distribution Tiny-ImageNet instead of in-distribution Fed-ISIC2019 as auxiliary dataset | BlindFed · Fed-ISIC2019 (federated K=5) | balanced accuracy: 0.6679 -> 0.6481 | Hurt |
| compare BlindFed to LoRA | LoRA · Fed-ISIC2019 (federated K=5, aux InD) | balanced accuracy: 0.6844 -> 0.6831 | Mixed |
| raise softmax Taylor degree from 2 to 6 (ablation) | FHE-friendly ViT · Fed-ISIC2019 softmax approximation | one-epoch accuracy (%): 75.17 -> 84.36 | Improved |
| raise softmax Taylor degree from 6 to 16 (ablation) | FHE-friendly ViT · Fed-ISIC2019 softmax approximation | one-epoch accuracy (%): 84.36 -> 85.08 | No clear effect |
Closest papers by meaning
- Data-Free Black-Box Federated Learning via Zeroth-Order Gradient Estimation (AAAI 2025 · similarity 0.86)
- Enhancing Federated Domain Adaptation with Multi-Domain Prototype-Based Federated Fine-Tuning (ICLR 2025 · similarity 0.86)
- Cluster Based Heterogeneous Federated Foundation Model Adaptation and Fine-Tuning (AAAI 2025 · similarity 0.85)
- Dual-Personalizing Adapter for Federated Foundation Models (NeurIPS 2024 · similarity 0.85)
- Federated Learning via Input-Output Collaborative Distillation (AAAI 2024 · similarity 0.85)
- FLUX: Efficient Descriptor-Driven Clustered Federated Learning under Arbitrary Distribution Shifts (NeurIPS 2025 · 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.
