Towards cardinal uncertainty quantification in AI
Posted on October 13, 2025 in SIPTA Seminar by Michele Caprio
Conformal prediction is an uncertainty representation technique. Given (i) a dataset with \(n\) elements sampled from a homogeneous enough population, (ii) a precision level \(\alpha\) between \(0\) and \(1\), and (iii) a function measuring how every element of the dataset “conforms/is similar to” the other ones, it delivers a conformal prediction region for the \((n+1)\)-th observation. The latter is contained in the conformal prediction region with probability at least \(1-\alpha\). Recently, conformal prediction has seen a surge in popularity given by its widespread adoption in Engineering and AI, two fields that are pervasive to modern-day STEM research.
Read more