MLOps means ‘machine learning operations’. It refers to a specific way of building and deploying machine learning (ML) systems that makes the process faster and more reliable. It does this by building on the principles of DevOps and modern software delivery to cultivate greater collaboration between the relevant individuals and teams.
Bringing machine learning systems into production can be complex; MLOps can make it easier for teams to deliver ML products quickly and consistently.
An approach to building and deploying machine learning solutions that leverages the techniques and practices of modern software engineering.
It makes it easier to deliver reliable machine learning products quickly.
MLOps is hard to do well. It also requires a very specific set of highly sought after skills.
It’s helping organizations address both internal demands for ML capabilities and embed ML in external-facing products.
What is it?
MLOps is both a set of practices and tools and a cultural mindset that, when combined, accelerate the development and deployment of machine learning solutions. It bridges the gap between the disciplines of machine learning, software engineering and data engineering, by outlining how these disciplines should collaborate with one another. This is important because data science work — often isolated and research-focused — is not traditionally equipped for the complexity of (often multiple) production environments.
MLOps is closely allied with something called CD4ML; however, the two things aren’t the same thing. MLOps refers to the overall philosophy and practice, while CD4ML is a technique within MLOps that uses continuous delivery tools and techniques to automate parts of the machine learning development process.
What’s in it for you?
Machine learning solutions are complex; it can be challenging to get them into production. This means organizations can miss out on the immense opportunities of the technology.
When used effectively it can cut the amount of time needed to deploy business-critical innovations. From business insights to new machine learning-driven product features, MLOps can ensure businesses get the most from machine learning.
What are the trade-offs?
MLOps isn’t easy to do. It requires a fairly sophisticated existing level of knowledge and expertise of machine learning systems for teams to do it effectively.
It also requires a very specific set of skills and knowledge that are in short supply. This means hiring people with MLOps experience or capabilities can be difficult (and probably expensive). It may also require changes to team structures to ensure people with relevant skills are working closely together.
How is it being used?
Most of us use some form of NLP every day to do things like online searches, spell-checking, and language translation. Enterprise use cases include customer service chatbots that deliver empathetic, personalized interactions based on the context of the conversation.
Conversational applications — like voice-activated virtual assistants used on smart speakers — rely on NLP to identify and contextualize speech and respond appropriately.
Elsewhere, NLP has been used to screen resumés to identify ideal candidates, and recruiters have even used NLP to craft stronger job descriptions that will appeal to the right applicants.
How is it being used?
Most of us use some form of NLP every day to do things like online searches, spell-checking, and language translation. Enterprise use cases include customer service chatbots that deliver empathetic, personalized interactions based on the context of the conversation.
Conversational applications — like voice-activated virtual assistants used on smart speakers — rely on NLP to identify and contextualize speech and respond appropriately.
Elsewhere, NLP has been used to screen resumés to identify ideal candidates, and recruiters have even used NLP to craft stronger job descriptions that will appeal to the right applicants.
