AI Paper Atlas

Research map / Federated learning and privacy / Privacy attacks and secure learning

FedSEA-LLaMA: A Secure, Efficient and Adaptive Federated Splitting Framework for Large Language Models

Zishuai Zhang, Hainan Zhang, Weihua Li, Qinnan Zhang, Jin Dong, Yongxin Tong, Zhiming Zheng

AAAI 2026 · Federated learning and privacy › Privacy attacks and secure learning · approach: Private federated aggregation · 1 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 collaborative KV cache (ablation)FedSEA-LLaMA · Query length 4000 inferenceinference speed: 8.2 times that without itImproved
add attention-mask compression (send seq_len, pad_len) (ablation)FedSEA-LLaMA · Distributed forward passtime per token: average time reduction of 87.6%Improved
use FedSEA-LLaMA instead of FL-GLM for distributed inferenceFL-GLM · Distributed inferencespeedup: over 7× speedupImproved
split model so client holds first/last blocks instead of full modelLLaMA2-7B · ReCoRD client GPU memoryGPU memory: 28.2G -> 3.4GImproved
increase noise scale from 0.0 to 0.1 (ablation)FedSEA-LLaMA Blocks_A=1 · MultiRCF1a/EM: 77.48/40.82 -> 59.11/8.18Hurt
increase noise scale from 0.0 to 0.1 (ablation)FedSEA-LLaMA Blocks_A=3 · MultiRCF1a/EM: 77.89/41.34 -> 75.68/34.10Hurt
increase Blocks_A from 1 to 3 at noise 0.1 (ablation)FedSEA-LLaMA · MultiRCF1a/EM: 59.11/8.18 -> 75.68/34.10Improved
increase noise scale from 0.0 to 0.02 (ablation)FedSEA-LLaMA Blocks_A=3 · MultiRCF1a/EM: 77.89/41.34 -> 78.61/42.39No clear effect
vary Blocks_A and Blocks_C from 1 to 3 (no noise) (ablation)FedSEA-LLaMA · MultiRCF1a/EM: 80.34/44.60 -> 81.17/47.22No clear effect
keep embedding plus client blocks instead of embedding only (ablation)FedSEA-LLaMA · ReCoRD model inversionBLEU-4 / Rouge-2: 0.051/0.003 vs FedBert 7.053 and 28.57Improved
add Gaussian noise 0.02 to hidden states (Blocks_A=3) (ablation)FedSEA-LLaMA · ReCoRD model inversionRouge-2 F1/BLEU-4: 0.069/0.016 -> 0.084/0.044No clear effect

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