Auto-FL-Research: Agentic Search for Federated Learning Algorithms
06:00 · July 3, 2026 · arXiv cs.AI RSS
Federated learning (FL) research often depends on many small but consequential algorithmic choices: optimizer variants, server aggregation rules, local training schedules, normalization, regularization, and model architecture. These choices are expensive to explore manually and difficult to compare fairly when candidate changes can also alter the FL training or evaluation path. In this work, we present Auto-FL-Research (AFR), a constrained coding-agent workflow for FL algorithmic recipe search. Agents may propose and implement candidate training algorithms, including server aggregation rules, client update schedules, local objectives, and registered model variants, while task profiles fix the mutation surface, compute budget, communication contract, and final model evaluation. Each campaign records candidate scores, runtime, edited files, artifacts, and failure status. We evaluate AFR on five healthcare cross-silo FLamby tasks and on grouped-client profiles for the five fixed LEAF datasets plus the LEAF synthetic task. Five-seed repeat evaluations support gains on four FLamby tasks and five of six LEAF profiles, while also exposing seed-sensitive and search-selected failure cases. Same-budget controls show that several gains correspond to FL-recipe changes, whereas other improvements are recovered by fixed-surface scalar controls or fail under repeat or held-out evaluation. These mixed outcomes are part of the contribution: they show how agent-generated candidates can be separated into repeated FL mechanisms, fixed-surface tuning effects, and selected single-run artifacts.
Summary
Federated learning research involves numerous interdependent algorithmic decisions, from server aggregation rules and client update schedules to local objectives, optimizers, normalization, and model variants. These choices interact with data heterogeneity and communication limits, making manual exploration costly and fair comparisons difficult when modifications can inadvertently alter training paths or evaluation protocols.
Auto-FL-Research (AFR) addresses this through a constrained coding-agent workflow built on NVIDIA FLARE. Task profiles define a fixed mutation surface, compute budget, communication contract, and evaluation harness, allowing agents to propose and implement code-level changes while preventing alterations to metrics, data splits, or the overall FL protocol. Each campaign logs candidate scores, runtimes, edited files, artifacts, and failure modes, with a review step that classifies outcomes as kept, discarded, or crashed. A literature-guided recovery loop activates when progress plateaus.
The framework was tested on five healthcare cross-silo tasks from FLamby and on grouped-client versions of the LEAF benchmarks, including the synthetic task. Five-seed repeat evaluations indicate improvements on four FLamby tasks and five of six LEAF profiles. Same-budget controls and held-out checks, however, reveal that some gains stem from repeated FL mechanisms, others from fixed-surface scalar tuning, and still others prove sensitive to seed or fail to generalize. The resulting records separate transferable algorithmic changes from single-run artifacts, providing a reproducible trace of what was attempted and which modifications survive rigorous validation.
Why it matters
Federated Learning is crucial for the Dutch AI market due to strict EU data privacy regulations (GDPR), especially in collaborative sectors like healthcare. This research provides advanced practitioners with an automated, agent-driven approach to optimize FL pipelines, directly supporting scalable and privacy-preserving AI development in the Netherlands.
