Smarter Everything with ML & the IoT
summary
Argues that doing inference on-device plus sending only predictions over cellular solves bandwidth, latency and privacy problems and demonstrates it end-to-end. Uses TensorFlow Lite for Microcontrollers and Edge Impulse to train a gesture classifier from ~12 minutes of accelerometer data; final model is ~19 KB flash, runs ~1 ms per inference and correctly detects idle/chop/wave. Uses a Blues Wireless Note module and Note Hub to send JSON predictions, with over-the-air model updates (DFU) that successfully push a new model to remote devices.
AI-written summary · about
chapters (10)
- 0:00Smarter IoT with ML
- 1:26IoT privacy concerns
- 5:16Move ML inferencing to edge
- 9:48Edge inferencing for maintenance
- 13:20TinyML benefits
- 17:34TinyML with Edge Impulse
- 31:10IoT connectivity with cellular
- 37:32Sending ML predictions to cloud
- 44:12OTA model updates
- 53:28IoT and ML in practice
related talks
01h 01mML and the IoT: Living on the Edge
59mML and the IoT: Living on the Edge
20mHow ML is Powering the Future of IoT
27mWhat’s TinyML good for
52mMachine Learning in IoT Applications
31mKeynote: The next big thing in machine learning is tiny
55mEdgeAI - AI in the Palm of Your Hand
32mAI for Mobile and IoT Devices: TensorFlow Lite