Research map / Federated learning and privacy / Personalized and heterogeneous FL
Developmental Federated Tuning: A Cognitive-Inspired Paradigm for Efficient LLM Adaptation
Yebo Wu, Jingguang Li, Zhijiang Guo, Li Li
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 |
|---|---|---|---|
| add deconfliction-guided layer grouping instead of random grouping (ablation) | DEVFT (LLaMA2-7B) · Close-ended average | average accuracy: 39.90 -> 42.33 | Improved |
| add deconfliction-guided layer grouping instead of even grouping (ablation) | DEVFT (LLaMA2-7B) · Close-ended average | average accuracy: 36.25 -> 42.33 | Improved |
| add deconfliction-guided layer grouping instead of random grouping (ablation) | DEVFT (LLaMA3.1-8B) · Close-ended average | average accuracy: 60.69 -> 64.25 | Improved |
| add deconfliction-guided layer grouping instead of even grouping (ablation) | DEVFT (LLaMA3.1-8B) · Close-ended average | average accuracy: 57.76 -> 64.25 | Improved |
| add differential-based layer fusion instead of random single-layer representative (ablation) | DEVFT (LLaMA2-7B) · Close-ended average | average accuracy: 37.72 -> 42.33 | Improved |
| add differential-based layer fusion instead of summing all group layers (ablation) | DEVFT (LLaMA2-7B) · Close-ended average | average accuracy: 40.90 -> 42.33 | Improved |
| add differential-based layer fusion instead of random single-layer representative (ablation) | DEVFT (LLaMA3.1-8B) · Close-ended average | average accuracy: 53.29 -> 64.25 | Improved |
| add differential-based layer fusion instead of summing all group layers (ablation) | DEVFT (LLaMA3.1-8B) · Close-ended average | average accuracy: 61.20 -> 64.25 | Improved |
| use initial submodel capacity 4 instead of 1 (ablation) | DEVFT (LLaMA3.1-8B) · Close-ended average | average accuracy: 61.56 -> 64.25 | Improved |
| use initial submodel capacity 4 instead of 32 (ablation) | DEVFT (LLaMA3.1-8B) · Close-ended average | average accuracy: 58.09 -> 64.25 | Improved |
| use growth rate 2 instead of 4 (ablation) | DEVFT (LLaMA2-13B) · Close-ended average | average accuracy: 46.47 -> 52.77 | Improved |
| use growth rate 2 instead of 8 (ablation) | DEVFT (LLaMA2-7B) · Close-ended average | average accuracy: 37.08 -> 42.33 | Improved |
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.
- ProgFed — ProgFed: Effective, Communication, and Computation Efficient Federated Learning by Progressive Training (ICML 2022) · better on 3 of 3 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
- Beyond Parameter Aggregation: Semantic Consensus for Federated Fine-Tuning of LLMs (NeurIPS 2026 · similarity 0.90)
- Ravan: Multi-Head Low-Rank Adaptation for Federated Fine-Tuning (NeurIPS 2025 · similarity 0.90)
- Federated Full-Parameter Tuning of Billion-Sized Language Models with Communication Cost under 18 Kilobytes (ICML 2024 · similarity 0.89)
- FedBPT: Efficient Federated Black-box Prompt Tuning for Large Language Models (ICML 2024 · similarity 0.88)
- Fed-SB: A Silver Bullet for Extreme Communication Efficiency and Performance in (Private) Federated LoRA Fine-Tuning (NeurIPS 2026 · similarity 0.88)
- Ferret: Federated Full-Parameter Tuning at Scale for Large Language Models (ICML 2025 · similarity 0.87)
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.
