Seminario
A Typed Natural Deduction System to verify Trustworthiness
Programma
A Synergia Research Seminars
Joint work with Fabio Aurelio D’Asaro
We present TPTND, a typed natural deduction calculus to verify trustworthiness of probabilistic programs as a safety property. To set the stage, we provide a simple case-study in a machine-learning setting to illustrate the need for such a system and its foreseeable use. For the formal presentation, we start with constructing distributions of aleatoric variables through formation rules, provide typing rules for deterministic programs under such distributions, and generalize to sampling under multiple instances of such programs to evaluate output frequency. We then formulate introduction and elimination rules for an operator to verify the trustworthiness of a program as an evaluation of its possibly unknown or opaque distribution against its fair and transparent counterpart. To conclude, we illustrate how standard meta-theoretical results of output preservation and progress for this calculus are interpreted to formulate a safety property.
Relatori/Relatrici
Giuseppe Primiero (University of Milan)
Dettagli sull'evento
Data e luogo
Inizio: 02/02/2022
alle ore 11:00
Fine: 02/02/2022
alle ore 13:00
Online su Zoom
Organizzato e promosso da:
Dipartimento di Scienze Pure e Applicate
Scuola di Scienze, Tecnologie e Filosofia dell'Informazione
LM-78 - Classe delle lauree magistrali in scienze filosofiche - Filosofia dell'informazione. Teoria e gestione della conoscenza
Synergia Research Group
Modalità di partecipazione
Altre informazioni utili
Per ulteriori informazioni: pierluigi.graziani@uniurb.it
