Implement FedCALM: Conflict-aware Layer-wise Mitigation for Selective Aggregation in Deeper Personalized Federated Learning (CVPR 2025)

[appleiphonedddd]

Implement the FedCALM method emphasizing conflict‑aware, layer‑wise mitigation and selective aggregation to improve deeper personalized FL models.

Goals

• Layer‑wise metrics/signals to detect aggregation conflicts (as defined in the paper).
• Selective or weighted aggregation per layer on the server.
• Support for deeper backbones and per‑client personalization components.

Scope

• PyTorch implementation with modular server/client roles.
• Hooks to collect per‑layer statistics (e.g., gradients/updates) required by the paper.
• Configs and scripts for experiments and ablations.

Proposed structure (suggested)

• fedcalm/
  • server.py, client.py

Tasks

• Implement client‑side training with hooks to compute/collect conflict‑related signals.
• Implement server‑side selective or weighted aggregation per layer based on these signals.
• Add tests for selection logic, aggregation correctness, and determinism where feasible.

Deliverables

• Clean, documented code with docstrings and comments.
• Reproduction scripts and logs/plots for main results and ablations.
• Config files and example command lines.

Acceptance criteria

• Stable training on at least one deeper model with non‑IID data.
• Improvements consistent with the paper’s claims (within reasonable tolerance).

References

• Paper: “FedCALM: Conflict‑aware Layer‑wise Mitigation for Selective Aggregation in Deeper Personalized Federated Learning (CVPR 2025)”
• Link(s): [FedCALM]
• Please cite the paper in README.

Labels
enhancement, help wanted, research, federated-learning, personalization, optimization