On-device training

Train and update machine learning models directly on the user's device (e.g., smartphone, IoT), ensuring data privacy and delivering real-time personalization.

On-device training enables model refinement using local data, eliminating the need to send sensitive user information to the cloud: this is privacy by design. The technology, powered by frameworks like LiteRT (formerly TensorFlow Lite), allows models to adapt to individual user behavior, improving personalization for tasks like fashion item recognition or predictive text. This process is critical for low-latency applications and for coping with model drift, ensuring the model remains accurate over time and can function without stable internet connectivity. TensorFlow Lite alone is active on billions of devices globally, proving the scalability and real-world utility of this edge-computing approach.

https://ai.google/edge/