Jenny Brown co-hosts with Mark Mirchandani this week for a great conversation about the ML lifecycle with our guests Craig Wiley and Dale Markowitz. Using a real-life example of bus cameras detecting potholes, Dale and Craig walk us through the steps of designing, building, implementing, and improving on a piece of machine learning software.

The first step, Craig tells us, is to identify the data collected and determine its viability in an ML model. He describes how to get the best data for your project and how to keep the data, code, and libraries consistent to allow better analysis by your ML models. He talks about the importance of a Feature Store to aid in data consistency. Craig explains how machine learning pipelines like TensorFlow are great tools to improve consistency in the ML environment as well, making it easier to improve your model and even to build new ones using the same data. Keeping this consistency from data scientist analyzation to ML developer to model deployment means a more efficient process and product.

Evaluating models after production is an important step in the lifecycle as well to ensure accuracy, validity, and performance of the model. Craig gives us some examples and tips on monitoring models after they’ve been deployed. We talk about the challenges of scaling ML projects and Craig offers advice for developers and companies looking to build ML projects.

Dale Markowitz

Dale Markowitz is an Applied AI Engineer for ML on Google Cloud. Before that, she was a software engineer in Google Research and an engineer at the online dating site OkCupid.

Craig Wiley

Craig is the Director of Product for Google Cloud’s AI Platform. Previous to Google, Craig spent nine years at Amazon, as the General Manager of Amazon SageMaker, AWS’ machine learning platform as well as in Amazon’s 3rd Party Seller Business. Craig has a deep belief in democratizing the power of data, he pushes to improve the tooling for experienced users while seeking to simplifying it for the growing set of less experienced users. Outside of work he enjoys spending time with his family, eating delicious meals, and enthusiastically struggling through small home improvement projects.

Cool things of the week

• Introducing GKE Autopilot: a revolution in managed Kubernetes blog
• At your service! With schedule-based autoscaling, VMs are at the ready blog

Interview

• Google Cloud AI and Machine Learning Products site
• GCP Podcast Episode 240: reCAPTCHA Enterprise with Kelly Anderson + Spring ML Potholes with Eric Clark podcast
• Using machine learning to improve road maintenance blog
• Key requirements for an MLOps foundation blog
• TensorFLow site
• Kubeflow Pipelines site
• TensorBoard site
• How to dub a video with AI video
• Can AI make a good baking recipe? video
• Machine learning without code in the browser video

What’s something cool you’re working on?

Jenny started a new podcast that reads interesting Google blog posts over at Google Cloud Reader.

Our friend Dr. Anton Chuvakin started the Cloud Security Podcast by Google. Read more about it and listen here. Follow the show and hosts on Twitter Cloud Security Podcast Anton and Tim

And listen to Anton on the GCP Podcast Episode 218: Chronicle Security with Dr. Anton Chuvakin and Ansh Patniak.