Information theory in Lean: the DPI
Lorenzo Luccioli (University of Bologna)
Abstract: The Data Processing Inequality (DPI) is a cornerstone of information theory, expressing the principle that information cannot be created through processing.
In this talk, I will present the formalization of information-theoretic results carried out by Rémy Degenne and myself using the Lean 4 theorem prover. The focus will be on the formalization of information divergences—particularly f-divergences—and their associated Data Processing Inequality (DPI), along with their connection to the framework of hypothesis testing. I will also discuss the key challenges and design decisions faced during the project.
logic in computer sciencemathematical softwareinformation theory
Audience: researchers in the discipline
( video )
Series comments: Most of the talks from the conference are available on YouTube: www.youtube.com/watch?v=ZPPDktjL1Lw&list=PLlF-CfQhukNlzXdQvu1SVt9vcD4--fLlg
| Organizers: | Jireh Loreaux*, Riccardo Brasca*, Kevin Buzzard* |
| *contact for this listing |
