The Department of Software Science invites applications for a postdoctoral researcher position in Diagrammatic Probabilistic First-order Logic for Safeguarding AI.
The goals of the position are to:
- perform high-level collaborative research in a team led by Prof Pawel Sobocinski;
- disseminate research results in top-tier conferences and journals;
- contribute to the ongoing research projects of the Estonian Center of Excellence in AI (EXAI);
- help to supervise graduate students in the Laboratory for Compositional Systems and Methods and
- the opportunity to contribute to teaching at undergraduate and postgraduate level.
CONTEXT:
The successful candidate will (i) participate in the activities of the Estonian Center of Excellence in AI and be actively involved it its ongoing research projects and (ii) work with a team in the Compositionality group at TalTech to introduce a combinatorial, programmatic, and diagrammatic syntax for probabilistic logic.
First-order logic has a recent string diagrammatic complete axiomatisation [1] in terms of first-order bicategories. At the same time, the morphisms of the Kleisli category of the finitary subdistribution monad have posetal structure given by a point-wise ordering. This structure is similar to the one of the regular fragment of first- order bicategories.
These observations point towards an extension of partial Markov categories [2] to a probabilistic analogue of cartesian bicategories of relations and first-order bicategories, with potentially richer 2-cells. Such structure could be further developed to incorporate traces, which would give a syntax for deriving statements for probabilistic processes analogous to Hoare triples for non-deterministic processes.
We intend this work to be a step towards an expressive formalism for the specification and reasoning about probabilistic systems, including systems that include AI compoments.
1. Filippo Bonchi, Alessandro Di Giorgio, Nathan Haydon, and Pawel Sobocinski. Diagrammatic algebra of first order logic. In 2023 39th Annual ACM/IEEE Symposium on Logic in Computer Science (LICS). ACM, 2024.
2. Elena Di Lavore and Mario Román. Evidential decision theory via partial Markov categories. In 2023 38th Annual ACM/IEEE Symposium on Logic in Computer Science (LICS), pages 1–14. IEEE, 2023.
Contact person: Pawel Sobocinski pawel.sobocinski@taltech.ee.