*Exciting PhD Opportunities in Artificial Intelligence Across Top Austrian Institutions*Are you passionate about tackling some of the most pressing challenges in Artificial Intelligence? My colleagues and I, from six partner institutions in Austria, are seeking talented and motivated PhD students to join us in cutting-edge research projects. We invite exceptional candidates to apply for fully funded PhD positions at JKU Linz, TU Wien, ISTA, TU Graz, AAU, and WU Wien. These positions are funded by the Cluster of Excellence (hashtag#CoE) “Bilateral Artificial Intelligence” (hashtag#BILAI) of Austrian Science Fund FWF.For more information, please visit: https://lnkd.in/dSvdYcUwI’m personally looking for a PhD student eager to dive into the development and applications of Description Logics and/or rule-based languages within the field of Machine Learning. Below are two examples of potential research projects you could be involved in:1. Knowledge Representation Techniques for Trustworthy Specification of Agents’ Rewards Reinforcement Learning (RL) has unlocked immense potential in building powerful AI systems. However, a crucial challenge lies in designing reward functions that guide AI agents. Reward functions are crafted in ad hoc ways, making it difficult to verify, explain, or ensure alignment with users' values and expectations. This project aims to develop new knowledge representation languages and reasoning techniques tailored for defining and inspecting reward functions. The goal is to create safer, more reliable, and transparent systems that avoid the errors and inconsistencies present in current methods. With logic-based techniques, we can allow automated reasoning to detect potential flaws and enable sharing and reuse of reward functions across different systems.2. Reconciling Graph Neural Networks and Two-variable Logics Recent advances in AI have introduced two powerful yet distinct methodologies for handling graph-structured data (such as Knowledge Graphs). On one hand, Graph Neural Networks (GNNs) offer machine learning-based techniques for prediction and classification on graphs. On the other hand, Two-Variable Logics (2VLs)—including formalisms like Description Logics (DLs), the W3C Web Ontology Language (OWL), and the W3C Shapes Constraint Language (SHACL)—provide logical frameworks for reasoning over graph-structured data. While GNNs and 2VLs share the same underlying graph model, they offer complementary strengths: 2VLs are explainable but computationally demanding, while GNNs are efficient but less transparent. The aim of this project is to combine the best of both worlds by creating hybrid frameworks that leverage the advantages of each, while mitigating their limitations. This work is in cooperation with Matthias Lanzinger.…more