Research map / Federated learning and privacy / Personalized and heterogeneous FL
Learning to Collaborate: An Orchestrated-Decentralized Framework for Peer-to-Peer LLM Federation
Inderjeet Singh, Eleonore Vissol-Gaudin, Andikan Otung, Motoyoshi Sekiya
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 |
|---|---|---|---|
| replace random P2P pairing with LinUCB matchmaking (CPM) (ablation) | P2P distillation · HumanEval+MBPP, 6 clients | Pass@1 (%): 8.89 -> 13.33 | Improved |
| replace data-diversity heuristic pairing with learned CPM (ablation) | P2P distillation · HumanEval+MBPP, 6 clients | Pass@1 (%): 6.67 -> 13.33 | Improved |
| pair clients by maximizing JS divergence instead of random (ablation) | P2P distillation · HumanEval+MBPP, 6 clients | Pass@1 (%): 8.89 -> 6.67 | Hurt |
| add P2P collaboration over isolated training | LLM agents · HumanEval+MBPP, 6 clients | Pass@1 (%): 2.22 -> 8.89 | Improved |
| use centralized averaged-logit KD instead of P2P | LLM agents · HumanEval+MBPP, 6 clients | Pass@1 (%): 18.33 -> 2.00 | Hurt |
| use FedID-CentralKD server teacher | LLM agents · HumanEval+MBPP, 6 clients | Pass@1 (%): 2.22 -> 1.11 | Hurt |
| use CPM-guided pairing instead of random (ablation) | P2P distillation · Transfer set peak student | Pass@1 (%): 33.33 -> 86.70 | Improved |
| use LinUCB CPM vs random pairing, high heterogeneity (ablation) | P2P distillation · Synthetic 32 clients | relative gain (%): 48.5 | Improved |
| use LinUCB CPM vs random pairing, low heterogeneity (ablation) | P2P distillation · Synthetic low heterogeneity | relative gain (%): 3.1 | No clear effect |
| heuristic pairing improves CodeBLEU over CPM (ablation) | Heuristic-P2P · HumanEval+MBPP, 6 clients | CodeBLEU: 0.344 -> 0.392 | Mixed |
Closest papers by meaning
- Federated Learning with Heterogeneous LLMs: Integrating Small Student Client Models with a Large Hungry Model (AAAI 2025 · similarity 0.87)
- Overcoming Data and Model heterogeneities in Decentralized Federated Learning via Synthetic Anchors (ICML 2024 · similarity 0.87)
- FedCompetitors: Harmonious Collaboration in Federated Learning with Competing Participants (AAAI 2024 · similarity 0.87)
- Helmsman: Autonomous Synthesis of Federated Learning Systems via Collaborative LLM Agents (ICLR 2026 · similarity 0.86)
- Free-Rider and Conflict Aware Collaboration Formation for Cross-Silo Federated Learning (NeurIPS 2024 · similarity 0.86)
- Bridging Model Heterogeneity in Federated Learning via Uncertainty-based Asymmetrical Reciprocity Learning (ICML 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.
