AI Paper Atlas

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

Bayesian Evidence-Driven Prototype Evolution for Federated Domain Adaptation

Xiaoyang Yi, Li Peng, Yuru Bao, Jian Zhang

ICLR 2026 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Federated learning with pretrained models

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 split rule R_split (ablation)FedPTE · Digit (5 clients)accuracy: 83.67 -> 85.72Improved
add split penalty P_split (ablation)FedPTE · Digit (5 clients)accuracy: 85.72 -> 86.31Improved
add merge rule R_merge (ablation)FedPTE · Digit (5 clients)accuracy: 86.31 -> 88.14Improved
add merge penalty P_merge (ablation)FedPTE · Digit (5 clients)accuracy: 88.14 -> 89.43Improved
add prototype topology-aware contrastive loss L_contra (ablation)FedPTE · Digit (5 clients)accuracy: 89.43 -> 93.98Improved
use NIW Bayesian evidence instead of random evolution (ablation)FedPTE · Office (4 clients)accuracy: 55.52 -> 73.01Improved
use NIW Bayesian evidence instead of SSE heuristic (ablation)FedPTE · Office (4 clients)accuracy: 62.89 -> 73.01Improved
raise contrastive weight lambda from 1 to 100 (ablation)FedPTE · Digit (5 clients)accuracy: 90.41 -> 93.98Improved
raise temperature tau from 0.02 to 0.06 (ablation)FedPTE · Digit (5 clients)accuracy: 93.11 -> 93.98Improved
vary NIW prior hyper-parameters kappa0 and nu0 (ablation)FedPTE · Digit (5 clients)accuracy: 93.63 to 93.98No clear effect
use adaptive instead of fixed split/merge penalties (ablation)FedPTE · Digit and Officeaccuracy: 92.65 -> 93.47 on MNIST-M; 71.79 -> 65.62 on DSLRMixed
add differential-privacy perturbation to prototypes (0.1 to 50.0) (ablation)FedPTE · Digit (5 clients)accuracy: 93.80 -> 78.44Hurt

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