Azure AI Workshop

Machine Learning Cloud NDC London

This workshop is part of NDC London 2019

With all the hype surrounding AI and machine learning, we are often left at a loss with the actual engineering aspects of this resurgent technology. This workshop is designed to help attendees understand the rudiments of machine learning through a guided set of hands on exercises in the field of computer vision. These problems will then be run at scale using the newly minted Azure Machine Learning service. Finally, the output of these machine learning techniques will be moved into actual software in a production environment. Attendees should come away with a solid understanding with the engineering process behind these amazing techniques that harness the power of data.

Agenda

  1. Machine Learning Overview
  2. Azure Machine Learning service overview
  3. Running Experiments
  4. Managing Models
  5. Intersecting Data Science with DevOps

Machine Learning Overview

  • Foundations of Machine Learning
    • Supervised vs. Unsupervised Learning
    • Features and Labels
    • Machine Learning Models
  • Pre-Built Models
    • Introduction to Cognitive Services
    • Computer Vision Service
    • Custom Vision Service
  • Deep Learning for Computer Vision
    • Linear Models
    • Multi-Layer Perceptron (Neural Networks)
    • Convolutional Neural Networks

Azure Machine Learning Service Overview

Logical vs. physical view of an AzureML service Workspace

  • Logical
    • Compute
    • Experiments
    • Data Stores
    • Models
    • Images
    • Deployments
  • Physical
    • Storage
    • Key Vault
    • App Insights
    • Azure Container Registry

Running Experiments

  • Moving execution to the cloud
  • Logging
  • Managing Data Stores
  • Experimentation Process
  • Hyper Parameter auto-tuning
  • Automatic Machine Learning

Managing Models

  • Saving Machine Learning Assets
  • Understanding ML Images

Models in Production

  • Model Deployment (ACI/k8s)
  • Models at the Edge

Azure DevOps

  • Using the deployment environment as an AI staging area
  • Managing API deployments in Azure DevOps with ACR
  • Automatic retraining of stale models

Tid

09:00 - 17:00