AI Paper Atlas

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

Investigating Social Bias Propagation in Federated Fine-tuning of Large Language Models

Jiaxu Zhao, Meng Fang, Mingze Zhong, Shunfeng Zheng, Ling Chen, Mykola Pechenizkiy

AAAI 2026 · Federated learning and privacy › Personalized and heterogeneous FL · approach: Non-IID federated learning · 0 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 BAMA bias-aware aggregationFedAvg, LLaMA 3 · 1B/2CCrowS-Pairs bias (dAvg): 21.06 -> 20.01Improved
add BAMA bias-aware aggregationFedAvg, LLaMA 3 · 1B/2CStereoSet bias (dAvg): 19.14 -> 17.96Improved
add BAMA bias-aware aggregationFedAdam, Phi 3 · 1B/2CCrowS-Pairs bias (dAvg): 18.03 -> 16.63Improved
add BAMA bias-aware aggregationFedAdam, Phi 3 · 1B/2CStereoSet bias (dAvg): 12.63 -> 10.42Improved
add BAMA bias-aware aggregationFedProx, Gemma 2 2B · 1B/2CCrowS-Pairs bias (dAvg): 17.44 -> 16.10Improved
add BAMA bias-aware aggregationFedAvg, Gemma 2 2B · 1B/2CCrowS-Pairs bias (dAvg): 17.52 -> 15.14Improved
use smaller model (Gemma 2 2B vs LLaMA 3 8B) with BAMABAMA · 1B/2CCrowS-Pairs bias (dAvg): 2.38 points vs 1.05 pointsImproved
use FedAdam instead of FedAvgfederated fine-tuning · Phi 3, 3CCrowS-Pairs bias (dAvg): 16.67 -> 13.12Improved
use FedProx instead of FedAdamfederated fine-tuning · LLaMA 3, Mix 1B/2CCrowS-Pairs bias (dAvg): 20.60 -> 21.36Hurt
use FedAdam instead of full fine-tuningLLaMA 3 · 3B biased dataCrowS-Pairs bias (dAvg): 21.22 -> 19.54Improved
use LoRA instead of full fine-tuningLLaMA 3 · 3B biased dataCrowS-Pairs bias (dAvg): 21.22 -> 20.84Improved
mix biased and clean data within each client instead of segregatingfederated fine-tuning · all models and FL algorithmsbias scores: 5%-7% higher biasHurt

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