Continuous delivery for machine learning — CD4ML — takes software engineering approaches and applies them to the creation of machine learning applications.
CD4ML promises to make the notoriously difficult task of deploying and improving machine learning applications less fraught. It can enable organizations to accelerate their efforts to become data driven and to maximize returns on their investments in machine learning. That can increase your time to market and your resilience in changing conditions.
What is it?
CD4ML applies continuous delivery practices to the deployment of machine learning applications. Continuous delivery is an approach to improving the software development process by getting updates to code into production quickly and regularly.
CD4ML is useful because while machine learning applications can be powerful tools, the process for developing, deploying and continuously improving them is complex. The applications’ behaviour is often hard to explain or predict; they’re hard to test and to improve.
With CD4ML, a cross-functional team produces machine learning applications based on code, data, and models in small and safe increments that can be reproduced, retrained, and reliably released at any time, in short adaptation cycles. The result is that your machine learning applications deliver business value faster.
What’s in for you?
Machine learning tools can have a profound impact on your ability to make business decisions based on data. CD4ML is a powerful approach to getting more business value from machine learning investments.
Whereas historically, machine learning applications have been somewhat opaque — in that they’re hard to test, explain, reproduce, and improve — CD4ML allows a more piecemeal deployment, that makes it easier to improve the applications throughout their lifecycle.
What are the trade offs?
CD4ML requires coordination between different disciplines, including data engineering, data science, testing, infrastructure engineering, and release engineering, while aligning this with business needs. For enterprises with siloed structures, this type of cross-functional cooperation may prove challenging.