Applied AI on the edge
summary
Many PoCs never reach production because streaming latency, cloud cost, privacy and unreliable networks block deployment. A usable path compiles ONNX model pipelines to WebAssembly binaries, ships a tiny 65 KB runtime to industrial gateways, and runs models locally (up to 12 fps for license‑plate OCR and servicing four cameras on ARM32). This reduces inference latency, cloud bills and data exfiltration, but engineering challenges remain: device diversity, cross‑compilation and predicting on‑device performance.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:53Why Scalable and edge AI
- 8:48Why AI projects fail
- 15:10Benefits of edge AI models
- 18:40Challenges at the edge
- 20:06Scalable for edge AI
- 23:54ONNX portable format
- 28:07Real-world apps and demos
- 31:41Summary
- 32:02Q&A
related talks
48mThe Power of Edge AI: Revolutionizing Future of Real-Time, Autonomous Processing
41mEscaping the Cloud: High-Performance AI in your Browser
07mLightning Talk: AI at the Edge: ONNX Inference in WASM on Featherweight K0s
33mMachine Learning in the Browser: Fast Iteration with ONNX & WebAssembly
33mBringing AI to the Heterogeneous Edge with WebAssembly & ONNX
42mAI Everywhere– accelerate your development from PC to cloud
10mThe Rise and Impact of Edge AI
21mUnlocking the potential of low-power embedded devices with Edge AI software