☁️🤖👁️ Deploying Cloud ML Models Using Microsoft InnerEye

[HChughtai]

☁️🤖👁️ Deploying Cloud ML Models Using Microsoft InnerEye

Project Leaders: Tom Dowrick (@tdowrick), Haroon Chughtai (@HChughtai)

Description

Microsoft InnerEye provides a suite of tools for training and developing medical imaging algorithms either using local computing resources or Azure cloud services, and there are many algorithms/models developed within CMIC that would benefit from compatibility with InnerEye, to allow for more widespread deployment.

During this hackathon project, participants will convert existing models developed within CMIC to use PyTorch Lightning, and then deploy them on the cloud using InnerEye. Participants will also work to deploy a local InnerEye environment using existing CMIC computing resources, to provide a testing/development environment that can be used in the longer term, and to prepare models before they are deployed on the cloud.

Ideal Participant Requirements

Not hard requirements - we can help with setup and familiarity during the hackathon

• Familiarity with Python, PyTorch, Git, machine learning, and working with remote compute.

Resources & Pre-Hackathon Setup

• Development Environment
  • Access to CMICHackathom AzureML Workspace (via UCL SSO - access to be provided by project leads)
  • Local machine setup for IE-DL with needed dependencies OR pre-setup AzureML compute instance (access provided by project leads)
  • Microsoft have provided $20,000 of Azure credits, and GPU access (8 x Tesla V100)​ 🙏🏽
• Inner Eye Background
  • Project InnerEye Open-Source Software for Medical Imaging AI
  • Inner Eye Deep Learning Repo (and the UCL Fork)
  • InnerEye-DeepLearning Documentation
• PyTorch Lightning

Hackathon Guidance & First Issues

1. Get access to AzureML Workspace
2. Train a Hello World segmentation model
3. Train a sample segmentation task - e.g. Segmentation of Lung CT
4. Bring Your Own PyTorch Lightning Model - Choose an existing segmentation model that exists in CMIC, convert to Pytorch Lightning, and train using InnerEye
5. Deploy your custom segmentation model

Outcomes

• Models deployed on Azure with InnerEye
• New functionality - PRs back to Microsoft repo
• The gaining of knowledge and having of fun.