Speaker: Luigi Quaranta, UNIBA

Over the last few years, the emergence of low-code ML frameworks and AutoML platforms has significantly democratized data-driven AI. Nowadays, these technologies allow any data owner – with little-to-no training – to easily build their own models and start leveraging ML-based predictions. However, integrating ML-based components in production systems is a complex endeavor stretching far beyond model training in the lab. To be deemed trustworthy, such components – and the development process thereof – need to meet several non-functional requirements (e.g., reproducibility, robustness, and scalability); this is especially true when models are targeted to safety-critical domains, such as medicine and security. The goal of this seminar is to provide an overview of the best practices and tools employed to date to manage the lifecycle of ML-based components, assuring the quality of resulting systems. Also, by increasing the awareness of the engineering challenges involved, this seminar aims to shorten the cultural gap between data scientists who lack a computer science background and software engineers.