Many mobile apps, e.g., virtual assistants, navigation apps, video apps, etc., use various machine learning (ML) models. Different features of the app may have respective associated ML models which may often not be available locally on a user device and need explicit user action to download from a server. This disclosure describes techniques for automatic synchronization of machine learning models to a user device. With user permission, ML model(s) of an app that are not available locally or for which a new version is available are automatically downloaded when certain conditions such as are met. The conditions can include network conditions and other device-specific parameters. The overheads for determining whether the conditions are satisfied are reduced by utilizing an on-device cache and flag, which can reduce the impact on app startup latency.

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.