ML and the IoT: Living on the Edge
summary
Shows a concrete pipeline: collect 29,000 48x48 labeled face images, train a convolutional network locally with Python/TensorFlow/Keras, hit ~66% validation accuracy, and offload heavy training to cloud GPUs while keeping development iterative. Demonstrates inference on CPU at ~2s per image versus Google Coral doing 7.6 ms / 130 fps and an Apollo dev board running keyword detection on a coin cell for $13 coin-cell demo. Covers Nvidia Jetson Nano, TensorFlow Lite / Lite Micro, mesh protocols (Thread, ZigBee) and federated/distributed learning possibilities.
AI-written summary · about
chapters (19)
- 0:00Start
- 5:23Machine Learning
- 6:16Deep Learning
- 7:08Explanation of Machine Learning
- 10:14The Cloud
- 11:42Model Training
- 12:40Emotion Classification Model
- 14:43Image Data Generator
- 18:35Inference
- 24:53Traditional Iot Installations
- 30:07Predictive Maintenance
- 32:10Face Tracking
- 33:22Iot Class Devices
- 34:49Nvidia Jetson Nano
- 41:17Emerging Technologies in the Iot
- 42:37Mesh Networking
- 52:31Speech Recognition
- 56:40Ota Firmware Updates
- 57:24Calls to Action
related talks
01h 01mML and the IoT: Living on the Edge
28mMachine Learning on the Edge
50mML and AI at the edge with Google Coral, EdgeTPU and Cloud Platform
49mEnable IoT with Edge Computing and Machine Learning
54mSmarter Everything with ML & the IoT
57mMachine Learning in the Age of IoT
01h 02mAI on the Edge
27mMachine Learning on the edge using TensorFlow Lite