From Models to Reality: Reliable Machine Learning for Environmental Monitoring and STEM

Marco Cardia

Universidad de Pisa, Italia

Resumen/Abstract

Artificial Intelligence is transforming how we interpret complex technical data, yet a persistent gap remains between high-performing models in controlled settings and their reliability in real-world applications. This seminar addresses this challenge by examining how Machine Learning systems can be designed, evaluated, and adapted to operate robustly in practice, focusing on two high-impact domains: environmental monitoring and STEM accessibility. The first part of the talk presents my research on AI-based "soft sensors" for real-time water quality assessment. We will discuss how UV-Vis spectroscopy, combined with ML, enables pollutant monitoring. The second part addresses my current work in STEM accessibility, evaluating the ability of state-of-the-art Multimodal Large Language Models (MLLMs) to describe technical diagrams and source code for blind and low-vision users.

Curriculum ponente

Marco Cardia is a PostDoctoral Researcher at the Department of Computer Science, University of Pisa, Italy. He holds a Ph.D. in Computer Science from the same institution, where his doctoral research is focused on the development of AI-based soft sensors for real-time environmental monitoring. His doctoral work, conducted in collaboration with the industrial research center ARCHA S.r.l., pioneered the use of generative models and deep learning for real-time water quality assessment using spectroscopic data. Currently, Marco is a Research Fellow focusing on the application of Generative AI and Human-Computer Interaction (HCI) to improve the accessibility of STEM content.