Introducing Continuous AI BKG

Introducing Continuous AI

June 29, 2021
by
· 4 min read

Those who watched the vision presentation at the recent AI Experience 2021 event might have noticed that we quietly announced a new MLOps feature called Continuous AI. I want to take some time and drill into this feature in a little more detail.

First off, I won’t call it a feature again. Calling something a feature diminishes its value, and Continuous AI is too important for that. It’s possibly the most innovative capability we have added to the DataRobot AI Cloud platform in some time.

Why do I think this? Well, just imagine a production machine learning model that always stays accurate after it’s deployed—all by itself. Now imagine predictions becoming even more accurate as time passes without unnecessary manual intervention.

Sounds pretty cool, right? 

This type of capability would have been handy in 2020 when the pandemic really kicked in and the lockdowns started. Patterns of the past meant nothing anymore. Machine learning models trained on 2019 data didn’t know what to do. They just panicked, and started barfing up bad predictions. Imagine trying to forecast the demand for Clorox wipes back in January 2020 when all you have to go on is quantity sold in the last month or in the same period last year. Good luck with that.

lockdown2
The above image shows how forecasting bus rides in Chicago became incredibly hard in March of 2020 when everyone suddenly started to work from home.

So, What Is Continuous AI?

Continuous AI is unique to DataRobot. With Continuous AI, you can create multiple MLOps retraining strategies to refresh your production models based on the schedule of your choosing—like when accuracy drops below a predetermined threshold or data drift occurs.

retraining
Above image shows a DataRobot MLOps retraining policy set to trigger when data drift occurs.

Continuous AI not only retrains your current production models for you. As part of the same process, it also generates and tests a whole host of new models and presents the top ones as recommended challengers. Challengers are then replayed against historical prediction data for you (or the system) to decide if one of them should be promoted as the new champion.

challengers
Above, Continuous AI has recommended 2 challenger models to potentially replace the current champion model

unnamed 7
Above, MLOps replays the challenger models on historical inference data to determine which model would have performed best.

Let me state again that this entire process can be fully automated. Imagine doing all of this manually—having to set up a process where every day, or even multiple times a day, your team rebuilds every production model for you using the newest set of data. Then the same team writes and tests hundreds of alternative models, using a mixture of traditional methods and the latest cutting-edge data science techniques and algorithms for every production model. Every day. Over and over.

You would need a team of thousands. And you would have to ask that team to stop sleeping. You’d have an employee retention issue.

So, what makes this possible? Well, Continuous AI does of course.

Continuous AI = AutoML + MLOps

Continuous AI combines the best of automated machine learning with the best of machine learning operations to continually improve models over their full lifecycle. 

I’m careful to use the term “lifecycle.”. The life of a model is cyclical. It’s not done after you first deploy it. Models learn from old data and make predictions. They must then re-learn from newer data to keep on predicting. This process needs to be constant. AI models don’t stop predicting until you retire them, and just like us mere mortals, they need to constantly learn to stay relevant. 

Ask Yourself This: If your model was accurate yesterday, will it still be accurate today?

Maybe or maybe not but the question is what are you doing today to mitigate this?

Using DataRobot MLOps, you can monitor, deploy, manage, and govern all of your production models from one place, regardless of how those models were built or where they are running. MLOps creates a Center of Excellence for your production AI models enterprise-wide and gives you the ability to track model health, accuracy, data drift, and a host of other metrics that tell you when your models need help. Not only does MLOps know when models need help, it also knows who (or rather what) to ask for help. It asks The Robot. It asks DataRobot AutoML.

Because DataRobot MLOps and AutoML are part of the same tightly integrated platform, when you combine the two, something quite magical happens. You get the ability to automatically build, operate, and improve the quality of all your production models, all of the predictions they generate, and, ultimately, all of the AI-powered decisions you make. And you can do this continuously and on autopilot—an evolving system, continually learning and constantly improving itself. 

Only DataRobot can do this and scale it across your entire organization. Continuous AI is a game-changer for organizations embracing data science and machine learning best practices.

pasted image 0 27
Continuous AI is a fundamental approach that takes AI projects with the DataRobot Platform to the next level.

Want to Learn More?

If I’ve done my job in this blog post, then you should want to learn more. Continuous AI can save you thousands of hours of productivity and prevent hundreds of thousands of bad predictions and even worse decisions. 

Check out the Continuous AI web page and watch the demo to learn more. You can also post a question in the DataRobot Community to have an expert weigh in. We can also perform a live demo, personalized just for you.

Thanks for reading and stay tuned for more exciting updates to DataRobot Continuous AI coming in 2021!

Demo
Are Your Models Ready for What’s About to Happen?
Learn More
About the author
Richard Tomlinson
Richard Tomlinson

Sr Director, Product Marketing, DataRobot

He works closely with product, marketing, and sales teams to drive adoption and enablement of data management and data engineering capabilities in the DataRobot AI platform. Richard has been working in the data warehouse, BI and analytics space for over 20 years with the last eight years focused on Hadoop and cloud platforms. He is based in Chicago but is originally from the UK and has a degree in statistics from the London School of Economics.

Meet Richard Tomlinson
  • Listen to the blog
     
  • Share this post
    Subscribe to DataRobot Blog
    Thank you

    We will contact you shortly

    Thank You!

    We’re almost there! These are the next steps:

    • Look out for an email from DataRobot with a subject line: Your Subscription Confirmation.
    • Click the confirmation link to approve your consent.
    • Done! You have now opted to receive communications about DataRobot’s products and services.

    Didn’t receive the email? Please make sure to check your spam or junk folders.

    Close

    Newsletter Subscription
    Subscribe to our Blog