Excited to share my new preprint:
Federated learning can be biased by who enrolls and who participates each round. Borrowing IPW from causal inference, we correct this two-stage bias and analyze how selection affects the optimization target and FL training dynamics.
arxiv.org/abs/2604.26604
arxiv.org
Who Trains Matters: Federated Learning under Enrollment and Participation Selection Biases
Federated learning (FL) trains a shared model from updates contributed by distributed clients, often implicitly assuming that contributing clients are representative of the target population. In pract...
