Quantifiers for Differentiable Logics in Rocq (Extended Abstract)
Research Output:
Conference Article in Proceeding or Book/Report chapter
Article in proceedings
Peer-reviewOpen access
Publication Information
Output type
Research Output:
Conference Article in Proceeding or Book/Report chapter
Article in proceedings
Peer-reviewOriginal language
EnglishPages from-to (Number of pages)
Pages 1-12 (12 pages)Publication milestones
- Published - 2025
Publication status
Published - 2025
Host publication title
8th International Symposium on AI Verification (SAIV 2025), Zagreb, Croatia, July 21--22, 2025Abstract
The interpretation of logical expressions into loss functions has given rise to so-called differentiable logics. They function as a bridge between formal logic and machine learning, offering a novel approach for property-driven training. The added expressiveness of these logics comes at the price of a more intricate semantics for first-order quantifiers. To ease their integration into machine-learning backends, we explore how to formalize semantics for first-order differentiable logics using the Mathematical Components library in the Rocq proof assistant. We seek to give rigorous semantics for quantifiers, verify their properties with respect to other logical connectives, as well as prove the soundness and completeness of the resulting logics.
Access to documents
Final published version
License:CC BY, opens in new tab
Related Event
Title
International Symposium on AI Verification
Event type
SymposiumDegree of recognition
International eventDate
21/07/2025 - 22/07/2025Location
ZagrebCroatia
