IA177 Neurosymbolic AI

Fakulta informatiky
jaro 2026

Předmět se v období jaro 2026 nevypisuje.

Rozsah
2/0/0. 2 kr. (plus ukončení). Ukončení: zk.
Vyučováno kontaktně
Vyučující
RNDr. Vít Musil, Ph.D. (přednášející)
Garance
RNDr. Vít Musil, Ph.D.
Katedra teorie programování – Fakulta informatiky
Kontaktní osoba: RNDr. Vít Musil, Ph.D.
Dodavatelské pracoviště: Katedra teorie programování – Fakulta informatiky
Předpoklady
Students should hold a bachelor’s degree in Computer Science or a related field, with a solid grounding in algorithms, formal languages, and discrete mathematics. Familiarity with propositional and first-order logic, probability, and linear algebra is assumed. No prior expertise in advanced machine learning is required.
Omezení zápisu do předmětu
Předmět je otevřen studentům libovolného oboru.
Anotace
This course provides a theoretical survey of Neurosymbolic AI, a field dedicated to synthesizing the strengths of classical symbolic reasoning and modern deep learning. It is an advanced course for students with a background in theoretical computer science, building upon foundational knowledge of logic, algorithms, and discrete mathematics. The emphasis will be on formal methods and algorithmic principles rather than practical implementation.
Výstupy z učení
After completing the course, a student will be able to: - explain the fundamental principles of symbolic reasoning and neural learning, and classify neurosymbolic approaches along different integration paradigms; - analyze and compare core methods such as differentiable logic, neural theorem proving, inductive logic programming, and graph-based reasoning; - apply formal models to evaluate how symbolic constraints can guide neural networks or how symbolic rules can be extracted from trained models; - critically assess neurosymbolic architectures in terms of scalability, interpretability, and theoretical guarantees;
Klíčová témata
- Foundations of symbolic reasoning and neural learning, their strengths, weaknesses, and motivation for integration - Differentiable logic and constraint-based learning (logic as a loss function, semantic regularization) - Neural theorem proving and differentiable inductive logic programming - Symbolic rule and program extraction for explainability - Graph neural networks for relational and knowledge-based reasoning - Differentiable programming and neural interpreters - Neurosymbolic reinforcement learning and hierarchical planning - Advanced topics: compositionality, causality, and world models - Applications in program synthesis, robotics, drug discovery, and vision-language tasks - Open research challenges and future directions in neurosymbolic AI
Přístupy, postupy a metody používané ve výuce
lectures, class discussion, reading
Způsob ověření výstupů z učení a požadavky na ukončení
oral examination
Vyučovací jazyk
Angličtina
Další komentáře
Předmět je vyučován každoročně.
Výuka probíhá každý týden.
  • Permalink: https://is.muni.cz/predmet/fi/jaro2026/IA177