Presentation: CI/CD for Machine Learning

Track: Machine Learning for Developers

Location: Ballroom A

Duration: 4:10pm - 5:00pm

Day of week: Wednesday

Share this on:

What You’ll Learn

  1. Learn why and how to build a CI/CD pipeline for ML models.
  2. Find out what are some of the tools to use in building a model’s CI/CD pipeline.

Abstract

Machine Learning is now widely used across our industry, yet we have very limited tooling when it comes to automating the ML model versioning, testing, and release. We will show how a CI/CD pipeline for ML can greatly improve both your productivity and the reliability of your software.

Question: 

What is the work you're doing today?

Answer: 

I work on the Azure DevOps, and I'm focused on integrations between external products and our products. It's helping people automate everything and go with where most of the demand is. One of the things that we're working on is helping people who are working on machine learning to automate some of their deployments which is still an unsolved use case in most cases.

Question: 

What goals you have with the talk?

Answer: 

I want to show the ability to build a full continuous integration and continuous delivery pipeline for machine learning models, the ability to start from train the model, and get all the way to production, and also walk through what amount really is and how we can simplify that process.

Question: 

What do you want people to leave the talk with?

Answer: 

If I had to summarize it in one line it would be any CI/CD pipeline is better than none. If you're going to automate major key pieces of this process will make your life a lot easier, simplify it and add speed to your deployments.

Question: 

Do you find that you typically need to deploy models?

Answer: 

Models actually are not that static and for big companies like us or Google or Facebook they change a lot even on a daily basis, but for smaller companies what I see right now is that it takes people months to actually deploy the models to production so they can't even change the model even if they wanted to.

Question: 

What are the tools for a model's CI/CD?

Answer: 

If you need to automate the training process, that piece can be filled by Azure Now. The market is still pretty young so there's just some tools coming up right now. Then you need the piece to put it all together in this CI/CD pipeline and deploy it, which is typically just deployment as an API endpoint, which you can consume as a service. For that you can use an automation tool such as, let's say Natural Pipelines or Jenkins. You can build those pieces out from the toolkits that are out there already, which allows you to get through from development to production in a couple of days instead of few months.

Speaker: Sasha Rosenbaum

Program Manager on the Azure DevOps Engineering Team @Microsoft

Sasha is a Program Manager on the Azure DevOps engineering team, focused on improving the alignment of the product with open source software.

Sasha is a co-organizer of the DevOps Days Chicago and the DeliveryConf conferences, and recently published a book on Serverless computing in Azure with .NET.

Find Sasha Rosenbaum at

Similar Talks

Machine Learning on Mobile and Edge Devices With TensorFlow Lite

Qcon

Developer Advocate for TensorFlow Lite @Google and Co-Author of TinyML

Daniel Situnayake

Evolution of Edge @Netflix

Qcon

Engineering Leader @Netflix

Vasily Vlasov

Mistakes and Discoveries While Cultivating Ownership

Qcon

Engineering Manager @Netflix in Cloud Infrastructure

Aaron Blohowiak

Monitoring and Tracing @Netflix Streaming Data Infrastructure

Qcon

Architect & Engineer in Real Time Data Infrastructure Team @Netflix

Allen Wang

Future of Data Engineering

Qcon

Distinguished Engineer @WePay

Chris Riccomini

Tracks

Monday, 11 November

Tuesday, 12 November

Wednesday, 13 November