Curriculum
Computer Science for Societal Challenges and Innovation, XXXVIII series
Grant sponsor
Università degli Studi di Padova
Supervisor

Fabio Aiolli
Co-supervisor
s
Konstantinos Priftis
Contact
luca.bergamin.3@studenti.unipd.it

Project description
Novel learning models, such as deep generative models, show wide applicability, from predictive maintenance applications to computational creativity, and can unlock a novel perspective on learning tasks. They are promising candidates for solving tasks that appear simple to humans but are very difficult to solve for machines, such as few-shot learning and out-of-distribution data processing. Deep generative models show the ability not only to discriminate between data points, but also to understand and replicate data-generating processes. Significant obstacles are still present, as their inner workings are often not transparent and are hardly able to be adjusted according to specific requirements. The lack of an effective way to instruct, control, and explain learning models hinders widespread adoption, trust, and innovation. I am interested in addressing such shortcomings and showing how we could conciliate a transparent structure with an effective model. To solve this challenge, pure deep learning-based systems are insufficient: we need to integrate high-level reasoning into AI. Using symbolic languages, existing work shows that we can perform logical inference and leverage prior knowledge within a neural architecture, but applicability to real-world tasks is still limited. My research is expected to impact the development of human-centered AI systems that have sustainability, fairness, and transparency as their focus.