AIoT Ops - Using Kubernetes and ML Ops to Build Edge ML Applications
Talkat KubeCon + CloudNativeCon North America 2021 · on Thursday, 2021-10-14· at 11:55· Room 502 AB + Online·
summary
Claim: a full ML Ops pipeline runs on the edge so training, drift detection, quantization and redeploy happen without cloud dependency. Implementation uses k3s, Argo Workflows and StreamZ with a custom Golang model/OTA registry. Training runs on an NVIDIA Jetson Nano, inference on Coral TPUs and an ESP32 + TensorFlow Lite MCU. Demo shows logistic regression trained and quantized, model uploaded, MCU reflashes over MQTT when a fault is detected, an LED blinks and an inference level triggers—the system works end to end.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:40What is AIoT
- 13:41Reference Architecture
- 20:31Reference Implementation
- 25:04Demo
- 28:06Workflow Pipelines
- 29:58Demo Setup
- 32:42Demo Walkthrough
- 35:36Model Registry
- 37:04Demonstration
- 40:34Questions
related talks
24mDemo of Kubernetes at the Edge for AI Model Deployments
10mReady, Set, Go for K8s at edge today for AI/ML with Tensorflow
20mHow ML is Powering the Future of IoT
29mTeam Teardown: MLOps in the Cloud for IoT
01h 01mML and the IoT: Living on the Edge
49mEnable IoT with Edge Computing and Machine Learning
21mMastering the DevOps Machine Learning Pipeline
52mMachine Learning in IoT Applications