Research map / Federated learning and privacy / Personalized and heterogeneous FL
FedP²EFT: Federated Learning to Personalize PEFT for Multilingual LLMs
Royson Lee, Minyoung Kim, Fady Rezk, Rui Li, Stylianos I. Venieris, Timothy Hospedales
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 federated learned Bayesian sparse rank selection (FedP²EFT) over FedL2P learning-rate personalization | FedL2P · MasakhaNEWS tir (seen clients) | accuracy (%): 45.3±0.7 -> 63.5±0.3 | Improved |
| add FedP²EFT over FedL2P | FedL2P · MasakhaNEWS xho (unseen clients) | accuracy (%): 64.4±0.9 -> 78.5±1.1 | Improved |
| add FedP²EFT over FedL2P on a FedDPA-T base model | FedL2P · XNLI ur (seen clients) | accuracy (%): 44.8±4.5 -> 63.7±0.2 | Improved |
| add FedP²EFT over FedL2P | FedL2P · MasakhaNEWS eng (unseen clients) | accuracy (%): 90.6±0.1 -> 90.7±0.1 | No clear effect |
| add FedP²EFT over LoRA | LoRA · Fed-Aya fr (unseen clients) | METEOR: 0.55 -> 0.22 | Hurt |
| add FedP²EFT over FedL2P | FedL2P · Fed-Aya ru (unseen clients) | METEOR: 0.25 -> 0.34 | Improved |
| use non-FL per-client rank selection (BT-LoRA, AdaLoRA) instead of fixed-rank LoRA (ablation) | LoRA · MasakhaNEWS and XNLI (seen clients) | accuracy (%): XNLI bg 45.8±0.3 -> 43.8±0.0 (BT-LoRA) | Hurt |
| use FedP²EFT instead of FedL2P for federated training of the personalization strategy generator | FedL2P · XNLI (mBERT) federated training | mean latency (s): 28.2 -> 3.4 | Improved |
| use FedP²EFT instead of FedL2P for federated training | FedL2P · Fed-Aya (Llama-3 3B) federated training | mean latency (s): 226.5 -> 80.3 | Improved |
| use FedP²EFT instead of FedL2P for federated training peak memory | FedL2P · Fed-Aya (Llama-3 3B) federated training | peak memory (GB): 32.2 -> 25.2 | Improved |
| use FedP²EFT instead of LoRA at inference | LoRA · Fed-Aya (Llama-3 3B) inference | peak memory (GB): 17.9 -> 15.7 | Improved |
| use FedP²EFT instead of LoRA at inference | LoRA · Fed-Aya (Llama-3 3B) inference | mean latency (s): 347.3 -> 400.6 | Hurt |
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.
- FedL2P — FedL2P: Federated Learning to Personalize (NeurIPS 2023) · better on 19 of 19 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
- Dual-Personalizing Adapter for Federated Foundation Models (NeurIPS 2024 · similarity 0.88)
- FedLLM-Bench: Realistic Benchmarks for Federated Learning of Large Language Models (NeurIPS 2024 · similarity 0.87)
- FlowerTune: A Cross-Domain Benchmark for Federated Fine-Tuning of Large Language Models (NeurIPS 2025 · similarity 0.87)
- Breaking Physical and Linguistic Borders: Multilingual Federated Prompt Tuning for Low-Resource Languages (ICLR 2024 · similarity 0.87)
- FedBPT: Efficient Federated Black-box Prompt Tuning for Large Language Models (ICML 2024 · similarity 0.87)
- Federated Fine-tuning of Large Language Models under Heterogeneous Tasks and Client Resources (NeurIPS 2024 · similarity 0.86)
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.
