Listen to your data: Swim builds AI models and predicts in real-time
summary
Swim automatically spins up stateful digital twins (tiny in-memory “web agents”) for each sensor, links neighbors, and runs streaming analytics and online learning at the edge; a demo targets traffic (Las Vegas ~60 TB/day, ~3,000 intersections) and claims ~10,000x latency improvement versus cloud. Each agent auto-dimensions a small DNN (three layers, random init) and backpropagates from live feedback, scales from Jetson/Arm devices to cloud GPUs, builds the app in ~600 lines, and streams predictions into Kafka/Azure.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:40Data is everywhere
- 3:23Digital twins
- 4:42Cloudscale problems
- 9:03Swim
- 12:45Stateful Concurrent Object
- 13:20In Section
- 15:05Compute
- 17:00Learning approach
- 18:55Application example
- 20:40Edge computing
- 21:50Web agents
- 23:40Observations
related talks
01h 56mEdge AI Systems
01h 07mPower of Possibility: Navigating the tech landscape
55mLarge Scale Distributed AI on the JVM
59mML and the IoT: Living on the Edge
21mA New Architecture for the Intelligent Edge
40mSmart cars of tomorrow: real-time driving patterns
42mSo, You Want to "Do Big Data?"
54mSmarter Everything with ML & the IoT