Research map / Federated learning and privacy / Personalized and heterogeneous FL
FedBABU: Toward Enhanced Representation for Federated Image Classification
Jaehoon Oh, SangMook Kim, Se-Young Yun
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 |
|---|---|---|---|
| Train and aggregate only the body while keeping the classifier fixed across clients | FedAvg · CIFAR100, 100 clients, s=10, f=0.1, τ=10 | personalized accuracy: 62.67 -> 66.32 | Improved |
| Train and aggregate only the body while keeping the classifier fixed across clients | FedAvg · CIFAR100, 100 clients, s=50, f=0.1, τ=10 | personalized accuracy: 38.09 -> 42.63 | Improved |
| Train and aggregate only the body while keeping the classifier fixed across clients | FedAvg · CIFAR100, 100 clients, s=100, f=0.1, τ=10 | personalized accuracy: 33.13 -> 35.94 | Improved |
| Apply body-only updates and aggregation to FedProx | FedProx · CIFAR100, 100 clients, s=10, f=0.1, τ=10 | personalized accuracy: 43.48 -> 67.15 | Improved |
| Use FedBABU rather than FedAvg | FedAvg · CIFAR100, s=10, f=0.1, τ=10; one fine-tuning epoch | personalized accuracy: 27.43 -> 63.29 | Improved |
| Use FedBABU rather than FedAvg | FedAvg · CIFAR10, 100 clients, s=10, f=0.1, τ=10 | personalized accuracy: 69.11 -> 76.25 | Improved |
| Fine-tune only the classifier during personalization (ablation) | FedBABU · CIFAR100, s=10, f=0.1, τ=10 | personalized accuracy: 64.77 -> 66.28 | Improved |
| Fine-tune the full model rather than only the classifier (ablation) | FedBABU · CIFAR100, s=10, f=0.1, τ=10 | personalized accuracy: 66.28 -> 66.32 | No clear effect |
| Fine-tune the full model rather than only the body (ablation) | FedAvg · CIFAR100, s=10, f=0.1, τ=10 | personalized accuracy: 27.94 -> 62.67 | Improved |
| Use the client class templates in place of the trained classifier for representation evaluation (ablation) | FedBABU · CIFAR100, s=10, f=0.1, τ=10 | initial accuracy: 18.50 -> 54.93 | Improved |
| Use the client class templates in place of the trained classifier for representation evaluation (ablation) | FedAvg · CIFAR100, s=10, f=0.1, τ=10 | initial accuracy: 14.40 -> 50.25 | Improved |
| Use body-only federation instead of full-model federation (ablation) | FedAvg with server-side shared data (p=0.10) · CIFAR100, s=10 | personalized accuracy: 49.62 -> 60.94 | Improved |
Later papers that compared with this method
16 later accepted papers report FedBABU as a baseline in their results tables; 16 of them report a better number on most of the results they share with it.
- Beyond Aggregation: Guiding Clients in Heterogeneous Federated Learning (ICLR 2026 · better on 4 of 4 shared results)
- FedPKDA: Personalized Federated Learning with Privacy-Preserving Knowledge Dynamic Alignment (AAAI 2026 · better on 4 of 4 shared results)
- FedARA: Resource-adaptive Low-rank Personalized Federated Learning via Anchor-driven Representation Alignment on Heterogeneous Edge Devices (CVPR 2026 · better on 1 of 1 shared result)
- Enhancing Visual Representation with Textual Semantics: Textual Semantics-Powered Prototypes for Heterogeneous Federated Learning (CVPR 2026 · better on 1 of 1 shared result)
- FedHB: Hierarchical Bayesian Federated Learning (JMLR 2025 · better on 13 of 13 shared results)
- PFedCS: A Personalized Federated Learning Method for Enhancing Collaboration among Similar Classifiers (AAAI 2025 · better on 3 of 3 shared results)
- Neural Architecture Search Driven by Locally Guided Diffusion for Personalized Federated Learning (ICCV 2025 · better on 2 of 2 shared results)
- pFedGPA: Diffusion-based Generative Parameter Aggregation for Personalized Federated Learning (AAAI 2025 · better on 1 of 1 shared result)
8 more not listed.
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
- On Bridging Generic and Personalized Federated Learning for Image Classification (ICLR 2022 · similarity 0.90)
- Personalized Federated Learning via Feature Distribution Adaptation (NeurIPS 2024 · similarity 0.89)
- Tackling Feature-Classifier Mismatch in Federated Learning via Prompt-Driven Feature Transformation (NeurIPS 2025 · similarity 0.88)
- pFedES: Generalized Proxy Feature Extractor Sharing for Model Heterogeneous Personalized Federated Learning (AAAI 2025 · similarity 0.87)
- Fine-Tuning Personalization in Federated Learning to Mitigate Adversarial Clients (NeurIPS 2024 · similarity 0.87)
- Flexible Sharpness-Aware Personalized Federated Learning (AAAI 2025 · 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.
