FIW WS 25/26

Human intelligence manifests itself through a number of highly specialized but interrelated competences and skills. Assuming that the basis for these abilities is an extraordinary well-organized associative memory, information and knowledge modeling takes on a special position for Strong Artificial Intelligence also called Artificial General Intelligence (AGI). A lot of commercial software systems use basic information modeling techniques, like Entity Relationship Diagrams (ERD) or Uniform Modelling Language (UML), which has strengths but also weaknesses. Some techniques are designed to model more dynamical events like the Business Process Modelling Language (BPML). Further developed modeling techniques are used to describe information in a more sophisticated way and even used to model knowledge represented in computer systems and AI systems. One example is the Knowledge Modeling and Description Language (KMDL). This course provides three main topics: a) The phenomenon of data, information and knowledge and advanced approaches to structure information for intelligent systems. b) The modeling of different aspects like time and space within that information models in different ways. c) Development of a custom modeling language providing an orthogonal integration of the discussed aspects. d) Implementing of parsers to compile the languages into adequate information and knowledge structures. The learned techniques are used to model the dynamic movement of a robot within a room with no coordinate system, but only based on relative and spontaneously received position information of detected items.

Welcome to the course Foundation of Neuro-Symbolic AI!

Modules content:
1. Logical Approaches [1]
- Semantics and syntax of Propositional Logic
- Logical inference algorithms
- Neural function decomposition and inference

2. Probabilistic Approaches [2]
- Independencies, Graphical Models and Sufficient Statistics
- Reasoning schemes and maximum likelihood estimation
- Causal models and inference

3. Statistical Models of Knowledge Graphs [3]
- Knowledge Graphs, Description Logic and Semantic Web Standards
- Query languages for inference
- Statistical Relational AI

4. Quantum Machine Learning [4]
- Circuit model of computation and measurement schemes
- Deutsch-Jozsa Algorithm and application in Neuro-Symbolic AI
- Quantum advantages in sampling and contractions

All concepts will be introduced in a unifying tensor network formalism [5], which will be developed sequentially during the course.
The topics are accompanied by demonstrations and exercises based on the python library tnreason.

Literature:
[1] Russell, Norvig: ArtificiaI Intelligence - A Modern Approach (fourth edition), Pearson Education 2021
[2] Koller, Friedman: Probabilistic Graphical Models: Principles and Techniques, MIT Press 2009
[3] Antoniou et al: A Semantic Web Primer (third edition), MIT Press 2012
[4] Schuld, Petruccione: Machine Learning with Quantum Computers (second edition), Springer 2021
[5] Goessmann: The Tensor-Network Approach to Efficient and Explainable AI, Technical Report on the ENEXA Project 2025

Strong artificial intelligence deals about the question how to design and build human like thought machines. In the course we elaborate the basics of human thinking including naming, numbering, abstracting, conceptualizing and what is needed to build a digital machine that is able to think and act like an individual and is able to interact with other individuals. A variety of interconnected and adopted concepts and theories are necessary to implement human like skills and behaviour. A selection of these, like theory of information, action, time and space as well as a theory of thinking is discussed and developed in the course.