ML and the IoT: Living on the Edge
summary
Shows how real projects move ML from cloud training to local inference: trains on a 28k-image Kaggle set, gets about 66% accuracy, and runs local CPU prediction in ~2s versus ~4s to Azure. Demonstrates edge hardware choices — a $150 Google Coral board hitting ~120 fps for face detection, and a $12 microcontroller running a ~18KB TensorFlow Lite model for gesture detection. Explains using TensorFlow Lite / TinyML for microcontrollers, federated learning, and why volume, latency and privacy push inference to the IoT.
AI-written summary · about
chapters (33)
- 0:00Intro
- 0:15Zeke
- 1:54Washing Machine
- 6:14ML vs Deep Learning
- 8:50The Streetlight Problem
- 11:51Human vs Machine Learning
- 12:53Discipline of Machine Learning
- 13:19Machine Learning Tools
- 13:49Example
- 19:38Train predict tune loop
- 21:11Train predict model
- 22:02Train predict classifier
- 23:50Deployment
- 24:38Pretrained APIs
- 25:50Azure Face API
- 27:21Cloud vs Local Performance
- 32:38The IoT is the source of data
- 34:28Volume
- 35:59The problem
- 36:58The solution
- 37:26What we want
- 38:05What does this look like
- 40:43What you get instead
- 41:24Google Coral
- 43:43Google Coral Demo
- 46:46IoT
- 47:27Raspberry Pie
- 48:28Lowcost ML
- 48:58Microcontrollers
- 56:29Federated Learning
- 58:05Three Big Things
- 58:18Single Board Computers
- 58:49Tensorflow Light
related talks
59mML and the IoT: Living on the Edge
54mSmarter Everything with ML & the IoT
49mEnable IoT with Edge Computing and Machine Learning
20mHow ML is Powering the Future of IoT
50mML and AI at the edge with Google Coral, EdgeTPU and Cloud Platform
55mUsing TensorFlow for Performing Deep Learning at the Edge
01h 53mTinyML: Machine Learning’s Future is Tiny and Bright
52mMachine Learning in IoT Applications