Intrinsic Training Signals for Federated Learning Aggregation
This provides a more efficient and seamless approach to federated learning aggregation for privacy-preserving collaborative training, establishing a new paradigm.
The paper tackles the problem of aggregating client-specific models in federated learning without architectural modifications by leveraging intrinsic training signals, achieving state-of-the-art performance on multiple benchmarks.
Federated Learning (FL) enables collaborative model training across distributed clients while preserving data privacy. While existing approaches for aggregating client-specific classification heads and adapted backbone parameters require architectural modifications or loss function changes, our method uniquely leverages intrinsic training signals already available during standard optimization. We present LIVAR (Layer Importance and VARiance-based merging), which introduces: i) a variance-weighted classifier aggregation scheme using naturally emergent feature statistics, and ii) an explainability-driven LoRA merging technique based on SHAP analysis of existing update parameter patterns. Without any architectural overhead, LIVAR achieves state-of-the-art performance on multiple benchmarks while maintaining seamless integration with existing FL methods. This work demonstrates that effective model merging can be achieved solely through existing training signals, establishing a new paradigm for efficient federated model aggregation. The code is available at https://github.com/aimagelab/fed-mammoth.