AI Paper Atlas

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

Understanding and Mitigating Dimensional Collapse in Federated Learning

Yujun Shi, Jian Liang, Wenqing Zhang, Chuhui Xue, Vincent Y. F. Tan, Song Bai

TPAMI 2023 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Non-IID federated learning · 92 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
add FedDecorr regularizerFedAvg · CIFAR10 (alpha=0.05)test accuracy (%): 64.85 -> 73.06Improved
add FedDecorr regularizerFedAvg · CIFAR100 (alpha=0.05)test accuracy (%): 59.87 -> 61.53Improved
add FedDecorr regularizerFedAvg · TinyImageNet (alpha=0.05)test accuracy (%): 35.02 -> 40.29Improved
add FedDecorr regularizerFedAvg · CIFAR10 (alpha=0.5)test accuracy (%): 89.84 -> 89.84No clear effect
add FedDecorr regularizerFedAvg · CIFAR100 (homogeneous, alpha=inf)test accuracy (%): 74.54 -> 73.87No clear effect
add FedDecorr regularizerFedAvg · TinyImageNet (homogeneous, alpha=inf)test accuracy (%): 49.33 -> 52.63Improved
increase number of clients to 100 with FedDecorrFedAvg · TinyImageNet (100 clients, alpha=0.05)test accuracy (%): 21.53 -> 30.55Improved
add FedDecorr on wider vs narrower networksFedAvg · CIFAR10 (alpha=0.05)test accuracy (%): ResNet18 71.51 -> 76.54; ResNet32 65.76 -> 67.21Mixed
use DeCov covariance regularization instead of FedDecorr correlation regularization (ablation)FedAvg · TinyImageNet (alpha=0.05)test accuracy (%): 35.02 -> 32.88Hurt
use Structured-DeCov instead of FedDecorr (ablation)FedAvg · TinyImageNet (alpha=0.05)test accuracy (%): 35.02 -> 32.04Hurt
add FedDecorr regularizerFedAvg · computation time TinyImageNettime per round (min): 25.4 -> 25.7No clear effect
vary local epochs E from 1 upward with FedDecorr (ablation)FedAvg · CIFAR100 (alpha=0.05)test accuracy (%): 50.67 -> 53.18Improved

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