The work “Things Machine Learning Models know that they don’t know” (link to be added asap) with Salvatore Ruggieri has been accepted at the Thirty-Ninth AAAI Conference on Artificial Intelligence (AAAI) 2025.
This paper collects our past experiences in the field of abstaining/deferring systems and provides a general overview of existing methods that aim to tackle the difficult task of building ML models that are self-aware of their own limits. Moreover, we envision a few research directions for future investigations in this field.
A final thanks to Salvatore Ruggieri for his support throughout this work and these past few years.