Versioning is the process of uniquely naming multiple iterations of a model that is created at different stages of ML development. It primarily helps in retracking a model to a previous iteration when faced with an adversary
Features are the data variables that are selected from the raw data to efficiently identify and solve the business problem using AI. Feature stores are repositories of such variables that can be used while building new AI models.
An ML pipeline usually refers to a process that orchestrates the end-to-end flow of data–raw data, features, model inputs, model outputs. Automating such a pipeline enables the extraction of data, processing it, and storing it at a data lake or warehouse before feeding into the model.
Continuous Integration (CI) enables teams to simultaneously work together and upload code, data, or features into a central repository. Continuous Delivery (CD) automates the deployment of the above elements into production by eliminating manual and multi-stage tasks.
Continuous Model Monitoring refers to keeping a watch on the models in production so that they stay consistent with their performance while delivering the set business objectives.
Some of the primary reasons behind the insufficient focus on monitoring are the lack of specialized knowledge, complexities with running ad-hoc scripts, and a time-consuming process due to irreproducible setup.