AI Paper Atlas

Research map / Federated learning and privacy / Communication-efficient FL

TCT: Convexifying Federated Learning using Bootstrapped Neural Tangent Kernels

Yaodong Yu, Alexander Wei, Sai Praneeth Karimireddy, Yi Ma, Michael Jordan

NeurIPS 2022 · Federated learning and privacy › Communication-efficient FL · approach: Federated optimization algorithms · 5 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
linear probing of last layer of FedAvg-trained modelFedAvg ResNet-18 · CIFAR-10 #C=2test accuracy: 56.9 -> 77.9Improved
retrain last layer from FedAvg features instead of random initResNet-18 · CIFAR-10test accuracy: 35.37 -> 77.93Improved
retrain last 6 layers from FedAvg features instead of random initResNet-18 · CIFAR-10test accuracy: 91.78 -> 91.91No clear effect
replace FedAvg with TCT (train-convexify-train)FedAvg · CIFAR-10 #C=2top-1 accuracy: 56.86 -> 83.02Improved
replace FedAvg with TCTFedAvg · CIFAR-100 alpha=0.001top-1 accuracy: 53.89 -> 68.42Improved
replace SCAFFOLD with TCTSCAFFOLD · CIFAR-100 alpha=0.5top-1 accuracy: 71.07 -> 69.68Hurt
extract eNTK from randomly initialized model (T1=0) instead of FedAvg-trained model (ablation)TCT · CIFAR-10/100test accuracyHurt
use SCAFFOLD gradient correction instead of FedAvg in Stage 2 (ablation)TCT · FMNIST, CIFAR-10, CIFAR-100training accuracyImproved
normalize eNTK features (ablation)TCT · CIFAR-10 #C=2training accuracy convergenceImproved
start from ImageNet pre-trained ResNet-18 instead of random bootstrapping (ablation)TCT · CIFAR-10 #C=2top-1 accuracy: 83.02 -> 92.97Improved
limit Stage 2 to one communication round (ablation)TCT · CIFAR-10 #C=2top-1 accuracy: 83.02 -> 64.94Hurt
use group norm instead of batch norm (ablation)FedAvg · CIFAR-10 #C=1top-1 accuracy: 11.27 -> 21.23Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

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

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