kapynResearch

A Practical Recipe for Semi-Supervised Federated ASR: Online Pseudo-Labels with Server Update Stabilization

A new paper tackles semi-supervised federated learning for automatic speech recognition. The authors identify two key design axes, the teacher model that generates pseudo-labels and the server-side anchor that stabilizes updates, as critical to preventing training divergence. Their recipe narrows the performance gap between semi-supervised and fully-supervised federated ASR, offering a practical path for leveraging unlabeled client data.

Apple ML Research·Sep 24, 2026

Opening Kapyn…