Explore data-driven applications and AIoT devices in this 32-minute conference talk from tinyML Asia 2020. Dive into model reduction techniques like network architecture search and adaptive quantization to optimize storage and computing efficiency while maintaining accuracy. Learn about a meta-learning approach for enhancing inference accuracy in diverse environments. Examine real-life examples in image and video services, and gain insights into future AIoT research directions. Discover how local intelligence solutions in IoT devices can improve data security and real-time performance under bandwidth constraints.