Scaling Client Participation in Federated LoRA Fine-Tuning for Multimodal Cross-Lingual Understanding

Abstract

Abstract: Federated Learning faces significant challenges due to data heterogeneity, particularly in non-independent and identically distributed (non-IID) environments. This issue hampers the accuracy and convergence of traditional aggregation methods like Federated Averaging (FedAvg). To overcome this problem, this study proposes FedMeanStd, a novel aggregation technique utilizing the mean and standard deviation of client updates to dynamically select participants. This selection ensures that only stable, relevant contributions are included in the global model, filtering out unreliable or extreme updat

Research Question

How does the scaling of client participation in federated LoRA fine-tuning influence convergence stability and final accuracy on multimodal cross-lingual understanding tasks?

Verification Level

Descriptive public verification status only; aggregate claim counts are public, but individual claim records are not exposed here.

Truth-Engine Gate Verdict

This record has not completed Gate 2 of the verification pipeline (a type-checked Lean4 proof for mathematical claims, or a sealed-sandbox reproduction for empirical claims). It is a literature synthesis only. VERIFIED requires an attached reproducible artifact (Lean4 proof source, or repro script and results) before this status can be set; it is not derived from review score or claim count.

Quality Tier

Descriptive public triage only; this tier does not alter current publication or DOI behavior.

Quality Dimensions

Automated triage signals derived from public fields; not human peer review or independent validation.

Correction Record

Public corrections are additive records. Current status does not claim the synthesis is error-free.

Provenance