STAI CDT student Fabrizio Russo presented the paper ‘Causal Discovery and Knowledge Injection for Contestable Neural Networks’ at ECAI 2023.
Fabrizio explains the paper and its impacts here; “We propose contestable neural networks: neural models whose outputs and rationale can be challenged by the people using them. We do this using (causal) graphs that represent the relationships that the model is using to come up with its predictions (outputs’ rationale) and allowing the domain experts building these models to inspect the computed graphs and modify them to align the model behaviour to their knowledge of the subject.Our method therefore increases model understanding, trust and improves accuracy leveraging upon expert knowledge.”
Fabrizio really enjoyed the conference and the opportunity to hear other research. He said, “The conference was great from the keynote to the technical and poster sessions. The social was a guided tour in the salt mines of Krakow, which was astonishing! We had the social dinner 134m underground! It was great fun.”
The full paper can be found here: Russo, Fabrizio & Toni, Francesca. (2023). Causal Discovery and Knowledge Injection for Contestable Neural Networks.