We have covered the conversion of models to run in the browser, and on iOS and Android devices. TensorFlow Lite can also run on the Raspberry Pi, a pocket-sized computer running Linux.
Moreover, devices designed specifically to run deep learning models started to emerge over the years. Here are a few examples:
- NVIDIA Jetson TX2: The size of a palm; it is often used for robotics applications.
- Google Edge TPU: A chip designed by Google for IoT applications. It is the size of a nail, and is available with a developer kit.
- Intel Neural Compute Stick: The size of a USB flash drive; it can be connected to any computer (including the Raspberry Pi) to improve its machine learning capabilities.
These devices all focus on maximizing computing power while minimizing power consumption. With each generation getting more powerful, the on-device ML field is moving extremely quickly, opening new applications every year.
