AI Paper Atlas

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

Collaborative Unsupervised Visual Representation Learning From Decentralized Data

Weiming Zhuang, Xin Gan, Yonggang Wen, Shuai Zhang, Shuai Yi

ICCV 2021 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Federated medical and multimodal learning · 98 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
aggregate and update the online encoder instead of aggregating the target encoder and updating the online encoder (ablation)FedU · CIFAR-10 non-IIDtop-1 accuracy (global predictor): 82.10 -> 84.07Improved
update only the target encoder with the aggregated encoder instead of the online encoder (ablation)FedU · CIFAR-10 non-IIDtop-1 accuracy (global predictor): 84.07 -> 9.99Hurt
update both online and target encoders instead of only the online encoder (ablation)FedU · CIFAR-10 non-IIDtop-1 accuracy (global predictor): 84.07 -> 81.24Hurt
update both encoders instead of only the online encoder with a local predictor (ablation)FedU · CIFAR-10 non-IIDtop-1 accuracy (local predictor): 82.18 -> 18.23Hurt
use divergence-aware predictor update instead of always local or always global predictor (ablation)FedU · CIFAR-10/100 non-IIDtop-1 accuracy: around 5%Improved
always update with the global predictor instead of local predictor (online encoder aggregation) (ablation)FedU · CIFAR-10 non-IIDtop-1 accuracy: 82.18 -> 84.07Improved
set DAPU threshold mu to 0.2 (ablation)FedU · CIFAR-10 non-IIDtop-1 accuracyImproved
increase local epochs E at fixed 500 total epochs (ablation)FedU · CIFAR-10 non-IIDtop-1 accuracyHurt
increase total training rounds R from 100 to 800 (ablation)FedU · CIFAR-10 non-IIDtop-1 accuracyImproved
use FedU instead of single client training (ResNet-50)BYOL · CIFAR-10 non-IID linear evaluationtop-1 accuracy: 77.84 -> 83.25Improved
use FedU instead of FedCA (ResNet-50)federated unsupervised learning · CIFAR-100 non-IID linear evaluationtop-1 accuracy: 42.34 -> 61.94Improved
use deeper backbone ResNet-50 instead of ResNet-18FedU · CIFAR-10 non-IID linear evaluationtop-1 accuracy: 78.71 -> 83.25Improved

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