Research map / Federated learning and privacy / Personalized and heterogeneous FL
Mitigating Non-IID Drift in Zeroth-Order Federated LLM Fine-Tuning with Transferable Sparsity
Yide Ran, Wentao Guo, Jingwei Sun, Yanzhou Pan, Xiaodong Yu, Hao Wang, Jianwen Xie, Yiran Chen, Denghui Zhang, Zhaozhuo Xu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use Meerkat 0.1% sparse ZO instead of full-parameter ZO | Full-FedZO · Non-IID avg accuracy, local step 10, LLaMA-3.2-1B | average test accuracy: 0.699 -> 0.759 | Improved |
| use Meerkat 0.1% sparse ZO instead of full-parameter ZO | Full-FedZO · Non-IID avg accuracy, local step 10, Qwen2-1.5b | average test accuracy: 0.761 -> 0.805 | Improved |
| use Meerkat 0.1% sparse ZO instead of full-parameter ZO | Full-FedZO · Non-IID avg accuracy, local step 10, Gemma2-2b | average test accuracy: 0.732 -> 0.772 | Improved |
| use Meerkat instead of Lora-FedZO at 100 local steps | Lora-FedZO · Non-IID avg accuracy, local step 100, LLaMA-3.2-1B | average test accuracy: 0.628 -> 0.726 | Improved |
| use Meerkat instead of Weight Magnitude sparsity on Gemma2-2b at 10 local steps | Weight Magnitude · Non-IID avg accuracy, local step 10, Gemma2-2b | average test accuracy: 0.779 -> 0.772 | Hurt |
| use Meerkat instead of Weight Magnitude sparsity on Gemma2-2b at 30 local steps | Weight Magnitude · Non-IID avg accuracy, local step 30, Gemma2-2b | average test accuracy: 0.770 -> 0.756 | Hurt |
| increase local steps from 10 to 100 in Meerkat (ablation) | Meerkat · Non-IID avg accuracy, LLaMA-3.2-1B | average test accuracy: 0.759 -> 0.726 | Hurt |
| reduce local step to 1 (high-frequency sync) in Meerkat (ablation) | Meerkat · Qwen2-1.5b, Dirichlet 0.5 vs IID | average test accuracy over seven tasks | Improved |
| add VPCS early stopping of extreme Non-IID clients (Meerkat-vp) (ablation) | Meerkat · Non-IID Dirichlet 0.5, avg over seven tasks | average test accuracy | Improved |
| random client selection for early stopping instead of GradIP-based selection (ablation) | Meerkat-vp · Non-IID Dirichlet 0.5, avg over seven tasks | average test accuracy | Hurt |
| extreme sparsity 1e-3 and 1e-4 in Meerkat at high-frequency communication (ablation) | Meerkat · T=1 communication | accuracy | No clear effect |
Closest papers by meaning
- Federated Full-Parameter Tuning of Billion-Sized Language Models with Communication Cost under 18 Kilobytes (ICML 2024 · similarity 0.92)
- Ferret: Federated Full-Parameter Tuning at Scale for Large Language Models (ICML 2025 · similarity 0.89)
- Beyond Parameter Aggregation: Semantic Consensus for Federated Fine-Tuning of LLMs (NeurIPS 2026 · similarity 0.89)
- Fed-SB: A Silver Bullet for Extreme Communication Efficiency and Performance in (Private) Federated LoRA Fine-Tuning (NeurIPS 2026 · similarity 0.88)
- Ravan: Multi-Head Low-Rank Adaptation for Federated Fine-Tuning (NeurIPS 2025 · similarity 0.88)
- Robust Federated Finetuning of LLMs via Alternating Optimization of LoRA (NeurIPS 2025 · similarity 0.88)
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.
