We are seeking highly motivated and talented individuals to join our dynamic research team for combining symbolic and sub-symbolic AI. The successful candidates will conduct research at the University of Klagenfurt in collaboration with our partner institutes ISTA, TU Graz, TU Vienna, and WU Vienna.
The vision of Bilateral AI is to educate a new generation of top-quality AI scientists with a holistic view on symbolic and sub-symbolic AI methods. The training will be distributed over the six participating universities. Joint seminars, scientific workshops, and compulsory courses outside the PhD students’ research fields, will also be designed to encourage interdisciplinarity. Apart from that, students will be involved in grant applications, conference organization, Bachelor and Master student supervisions, and teaching. Each student will be supervised by two experienced and internationally renowned professors with different research fields (symbolic / sub-symbolic AI). The training will also provide a career development program, advice and support for students with innovative business ideas, and workshops for presentation and soft skills.
Open until filled. Applications will be processed regularly.
Only full application documents will be considered submitted as detailed below.
Our project is committed to increase the proportion of academic female faculty and, for this reason, especially welcomes applications by qualified women. If applicants are equally qualified, a woman will be given preference for this position.
How to apply:
If you are interested in a position, please send your regular application documents including
via email to: bilateral-ai-recruiting@aau.at
In addition to the application documents, two reference letters shall be sent with the subject "Candidate name_of_the_candidate" to bilateral-ai-recruiting@aau.at within two weeks from submission of the application.
Personalized recommendations have become a pervasive element of today's online experience, appearing on social networks, e-commerce sites, and media streaming services. Modern recommender systems are commonly based on machine learning algorithms, which suggest items based on behavioral patterns derived from large amounts of data. While machine learning algorithms can be highly effective, they also have certain limitations. For example, they often act as black boxes, leading to limited transparency and explainability. Additionally, the learning process may introduce biases, resulting in recommendations that may be considered inappropriate. This project addresses these problems by exploring the incorporation of additional forms of knowledge into the recommendation processes. Specific problem areas to be addressed in the project may include algorithmic biases, questions of fairness, the provision explanations, and cold-start issues.
Machine learning and deep neural networks are successfully applied to reason about massive amounts of available data, while explainability and trustworthiness of the outcomes are limited. Knowledge representation and reasoning methods, on the other hand, are geared for automated problem-solving and can provide verifiably optimal solutions for computationally complex tasks. The current challenge lies in bringing these orthogonal reasoning approaches together by transferring data-driven information into knowledge as well as utilizing knowledge to draw more meaningful and reliable conclusions from data. The goal of the project is to design hybrid approaches combining data- and knowledge-driven reasoning, with particular focus on realistic application areas like healthcare assistance, traffic monitoring, and green AI.
Sometimes trained traditional neural networks are binarized for improved efficiency, and more recently fully binary neural networks have been successfully employed as well. This development can be seen as facilitating interactions between sub-symbolic and symbolic reasoning because at least large parts of these networks are logical circuits and hence logical formulas. Some work has been done in this area already but seems to focus on the verification of properties of BNNs. In this project, we will try to establish which logics and logical methods best model BNNs, which methods from logic can be profitably used, most notably for performance improvements, and how various decision modes like nonmonotonicity are achieved in BNNs. Consequently, there are theoretical and practical aspects to be addressed.
A central grand vision of AI is that “the user states the problem; the computer solves it.” It aims at a broad application of AI for problem-solving without human assistance. However, in most application cases, successfully employing AI problem-solving methods requires considerable human intellect and effort to implement application-specific problem-solvers. The main hurdles are excessive memory consumption and unfeasible runtime, even for tasks humans can easily solve by employing decompositions, heuristics, and abstractions. Consequently, we aim to learn such strategies and their incorporation into current problem-solving methods (e.g., logic programming/answer set programming or constraint programming) to move from narrow to broad AI where computers solve problems without human assistance. For this project, we seek two PhD students working in a team.