AI Paper Atlas

Research map / Federated learning and privacy / Personalized and heterogeneous FL

Navigating Data Heterogeneity in Federated Learning: A Semi-Supervised Federated Object Detection

Taehyeon Kim, Eric Lin, Junu Lee, Christian Lau, Vaikkunth Mugunthan

NeurIPS 2023 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Federated semi-supervised and graph learning · 6 citations (OpenAlex)

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 changedWhereMeasuredEffect
Use FedSTO instead of FedAvgYOLOv5 Large SSFOD · BDD100K Non-IIDmAP@0.5: 0.558 -> 0.593Improved
Use FedSTO instead of FedAvgYOLOv5 Large SSFOD · BDD100K IIDmAP@0.5: 0.591 -> 0.609Improved
Add local EMA pseudo-labeling to federated training (ablation)SSFOD · BDD100K Non-IIDmAP@0.5: 0.511 -> 0.558Improved
Add selective training after SSFL with Local EMA Model (ablation)SSFOD · BDD100K Non-IIDmAP@0.5: 0.558 -> 0.567Improved
Add selective training after SSFL with Local EMA Model (ablation)SSFOD · BDD100K IIDmAP@0.5: 0.591 -> 0.587Hurt
Add full-parameter training with orthogonal enhancement (ablation)SSFOD with selective training · BDD100K Non-IIDmAP@0.5: 0.567 -> 0.593Improved
Add full-parameter training with orthogonal enhancement (ablation)SSFOD with selective training · BDD100K IIDmAP@0.5: 0.587 -> 0.609Improved
Remove selective training (ablation)FedSTO · BDD100K Non-IIDmAP@0.75: 0.289 vs 0.338Hurt
Remove selective training (ablation)FedSTO · BDD100K IIDmAP@0.75: 0.311 vs 0.357Hurt
Use client-local EMA instead of a global pseudo-label modelFederated SSFOD · BDD100K Non-IID, weather-specificmAP@0.5: 0.555 -> 0.560; 0.560 -> 0.566; 0.497 -> 0.553; 0.488 -> 0.553Improved
Use a reduced Phase 1 model that excludes the neck when communicatingFedSTO · 100 clients, 350 roundsnetwork bandwidth: 2,725.50 GB -> 2,166.23 GB; 20.52%Improved
Increase client sampling ratio from 0.1 to 0.2FedSTO · BDD100K, unlabeled BusmAP@0.5: 0.573 -> 0.378Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy