Curriculum
Computer Science and Innovation for Societal Challenges, XXXVI series
Supervisor
s
Gian Antonio Susto
Contact
alessio.arcudi@phd.unipd.it

Project description
ML and Anomaly Detection solutions provide automatic and intelligent tools that lead to waste prevention and optimized use of resources, but a lack of understanding of these black-box tools can have a negative effect on various aspects, like trust and adoption of the model, improvement and robustness in critical conditions. Research interests: Explanations of models. Strengthening interpretability by introducing explicit constraints on hidden representations on architecture can shed light on the reliability of black-box models, as well as directly include humans in the optimization cycle, leveraging their feedback to improve interpretability of the models. The explanations not only have a good effect on the reliability of the ML models, but also provide information on the causes that lead to an anomaly in the process.