Maintainability of AI Systems

This course is officially recognised and labelled by the European Institute of Innovation and Technology (EIT). EIT Label is a quality mark awarded to programmes demonstrating outstanding innovation, educational excellence and societal impact.

A data science pipeline refers to the process of collecting, cleaning, processing, analyzing, and visualizing data to extract insights and make informed decisions. It involves a series of stages that can vary depending on the specific project, but generally, the last step involves deployment of the models into production, that is to say, make them available to users for taking real-world decisions. After deploying the machine learning models, there are multiple challenges that can be faced. These challenges, if not handled properly, will deteriorate the performance of the models, thus making the AI solution of very short-life time.

Monitoring is a critical property of the data science pipeline, that makes it maintainable. A well-designed pipeline should be easily maintainable, meaning it should be easy to update and modify as new data sources become available or new business requirements emerge. Essential best practices for maintaining a data science pipeline will be presented in this learning path, including 1) benefiting from a modular design of the pipeline, where each stage is self-contained and can be easily updated or modified without affecting the other stages, 2) utilizing monitoring and alerting systems to detect any issues or errors in the pipeline, while keeping the AI solution robust against data drifting,

By implementing these best practices, a data science pipeline can be easily maintained, updated, and modified, ensuring that it continues to deliver accurate and relevant insights to the organization.