Model deployment is the final step in the machine learning process before a model can make predictions in the real world. These predictions will carry real consequences when forming the basis of decisions.
Even though model deployment is not typically the data scientist's job, aspects of this process are good to know when building a case and informing your questions.
Deployment is the process by which a machine learning model is added to existing infrastructure that powers a platform in the real world — like a website, app, or another digital interface.
Scalability is a real issue for many machine-learning projects. An analyst needs to make sure that their models will scale and meet increases in performance and application demand in production.
Data scientists usually rely on relatively static data on a manageable scale at the beginning of a project. As the model moves forward to production, it's exposed to larger quantities of data. The team working on deploying the model will need several tools to monitor and solve the performance and scalability challenges that will show up over time.
Something else to consider is that when deployed, many machine-learning algorithms are not run occasionally but continuously. This requires writing programs that regularly feed new data into the trained algorithm and building back-end data infrastructure.
Machine-learning algorithms running at scale may also be turned into online learning systems. This means that algorithms are regularly and automatically re-trained triggered by the collection of new data. These also require additional back-end infrastructure.
For the algorithm to be usable in the real world, it needs a user interface. Those working in the welfare, criminal justice, immigration, and policing systems are not data scientists and do not know how to interact with algorithms on that level. It is the team's responsibility to develop a user-friendly front-end to deploy this algorithm fully.
From there, the model is deployed — the machine-learning algorithm can make predictions and decisions in the real world.
At this point, we hope you have a deeper understanding of what critical decisions drastically affect the way an algorithm will categorize or predict an outcome. We hope that with this overview, you'll be able to ask more detailed questions, do adequate research, and strengthen your litigation case. If you have any questions, comments, or suggestions, please share them with us through the contact form!
If you notice anything incorrect or missing from our explanations, please let us know through the contact form below! We want our content to be as accurate and useful as possible — and you can help us do that.