How to set up federated fine‑tuning of a customer‑support model on user devices with privacy guarantees using flower and differential privacy
I recently set up a proof‑of‑concept to fine‑tune a customer‑support text classifier across users' devices, with the twin goals of keeping raw messages on device and giving users formal privacy guarantees. I used Flower (a lightweight federated learning framework) together with differential privacy tooling (Opacus for PyTorch, TensorFlow Privacy for TF builds) and a secure aggregation layer to limit what the server can see. Below I walk...