Edge ML Architectures

This post walks through four edge architecture design patterns for running your ML models at the edge.

Edge ML Architectures

Edge deployment refers to the deployment of machine learning (ML) models on devices at the edge of a network. Running ML models at the edge enables real-time predictions, lower latency, and increased security, but also presents unique architectural challenges.

In this webinar, we explore different paradigms for deploying ML models at the edge, including cloud-edge hybrid architectures and standalone edge models. We cover why device dependencies like power consumption and network connectivity make setting up and running ML models on edge devices chaos today, and discuss the elements needed for an ideal edge architecture and the benefits of this approach.

In this video, we walk through four edge ML architectures: 

  • Native edge
  • Network-local
  • Edge cloud
  • Remote batch

... and also show three demos to help you see how these design patterns power real ML-enabled solutions running at the edge. You'll see an edge-centric NLP web app, defect detection at the edge, and computer vision running in parking lots. Join us as we go out on the edge of glory to learn more about an edge-centric approach to ML deployments.

Video Breakdown

Want a breakdown of what we cover? Skip ahead to:

06:00 Why Run ML at the Edge? and Edge Defined

08:56 Device dependencies: power consumption and network connectivity

12:05 Factors that make setting up one device today...chaos!

17:18 Elements for an ideal edge architecture

21:20 Recipe for building an edge-centric architecture

23:57 Benefits of an edge-centric architecture

25:35 Demo 1: Edge-centric NLP web app

32:41 Edge-first design pattern: Native edge

35:11 Edge-first design pattern: Network-local

37:46 Edge-first design pattern: Edge cloud

40:13 Edge-first design pattern: Remote batch

41:07 Demo 2: Defect detection at the edge

47:45 Demo 3: Computer vision parking lots

53:39 Conclusion: the edge of glory