Overview

The Institute of Information Systems at the university of Lübeck together with the Faculty of Information Systems and Applied Computer Sciences (Professorship Smart Environments) at the  University of Bamberg offer various co-advised themes for master theses dealing with recent exciting research questions in the context of knowledge graph embeddings (KGE).  KGE can be seen as link between machine learning (MK) and knowledge representation. In KGEs, concepts and relations are represented by geometric structures that are obtained by ML techniques. The explicit representation of concepts and relations empowers a form of reasoning that augments the ML technique. Cones have been identified  as a simple yet rich class of geometrical structures for KGE. [OLW20,BYRL21,ZWJSD21]. In particular, cones allow to embed ontologies with full (ortho-)negation [G74].

Topics

Here are some topics in the area sketched in the overview. For a master thesis in one of the following or related topics you may contact the potential advisors at the University of Bamberg or the University of Lübeck mentioned below.

Topic: Ontology-informed Zero-Shot Learning with cone embeddings

The master thesis is going to deal with implementing an embedding approach with the specific task of Zero-Shot Learning (ZSL). ZSL is a multi-class learning task in which each instance has to be assigned exactly one label. [XSA17]

The distinct feature of ZSL is that instances to be classified by a label not seen during training. To be able to label previously unseen classes, some auxiliary information is needed. This can be done in form of per-class-attribute information, meaning for each class a set of attributes is given and specified whether one is positive or negative. The approach is intended to lead to more accurate results by taking into account explicit statements relying on assertions relying on negation [G21,G74] and relying on the polarity operator over cones for that purpose.  

Topic: Logical commitments of knowledge graph embedding approaches

This master thesis is placed in the wider context of explainable AI (XAI) which aims explaining (under a very wide reading) decisions an intelligent agent takes in a human comprehensible way. [L19] In particular, the master thesis is going to contribute to this area by a systematic analysis, comparison and survey of SOTA embedding approaches by identifying their logical commitments.   
Logical commitment can be understood as an advancement of ontological commitments used in context of discussing knowledge representations [DSS9393, G95]. The idea behind this advancement is is to extract the (symbolic) reasoning capabilities of KGEs. And to do so, one has to identify the language to compose  statements and the calculus to draw conclusions—put differently, one has to identify the logic. The class of orthologics (and their extensions)  give a nice framework to pursue a comparison in particular for those approaches which provide a negation [G21,G74].   

Topic: Knowledge graph embedding with reification

 This master thesis is going to deal with an an embedding approach using reification [Q53,MK01].  Reification is a general methodology being discussed in (the philosophy of) logics [Q53, MK01] but also known and used in  databases design as well  knowledge representation and the semantic web [W3C].    Its idea is to represent linguistic entities (such as sentences, or parts of them such as predicates) as objects in the embedding space. In the case of knowledge graph embeddings, the master thesis will consider reifying triples (a R b). Such a triple asserts that the object a is R-related to object b. Its reification would lead to an object c and statements that c ist related to object a via a function (for first argument) and via a function (for second) to object b. By this form of reification the general problem for learning arbitrary relations is reduced to that of learning functions. This is a tremendous gain in in theory, but has to be verified experimentally with a to-be-developed prototype.    
  

References

• [BYRL21] Y. Bai, R. Ying, H. Ren, and J. Leskovec. Modeling Heterogeneous Hierarchies with Relation-specific Hyperbolic Cones. In Proc. 35th Annual Conference on Neural Information Processing Systems (NeurIPS 2021), page arXiv:2110.14923, 2021.

• [OLW20] O. L. Ozcep, M. Leemhuis, and D. Wolter. Cone semantics for logics with negation. In C. Bessiere, editor, Proc. of IJCAI 2020, pages 1820–1826. ijcai.org, 2020.

• [LOW20] M. Leemhuis, O. L. Ozcep, and D. Wolter. Multi-label learning with a cone-based geometric model. In M. Alam, T. Braun, and B. Yun, editors, Proceedings of the 25th International Conference on Conceptual Structures (ICCS 2020), pages 177–185, 2020.

• [ZWJSD21] Z. Zhang, J. Wang, C. Jiajun, J. Shuiwang, and W. Feng. Cone: Cone embeddings for multi-hop reasoning over knowledge graphs. In Advances in Neural Information Processing Systems, 2021.

• [G74] R. I. Goldblatt. Semantic analysis of orthologic. Journal of Philosophical Logic, 3(1):19–35, 1974

• [XSA17]Y. Xian, B. Schiele, and Z. Akata. Zero-shot learning - the good, the bad and the ugly. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), July 2017.

• [L19] F. Lecue. On the role of knowledge graphs in explainable ai. Semantic Web Journal, (2198-3411), 2019.

• [DSS93] R. Davis, H. Shrobe, , and P. Szolovits. What is a knowledge representation? AI Magazine, 14(1):17–33, 1993.

• [G95] T. R. Gruber. Toward principles for the design of ontologies used for knowledge sharing. International Journal of Human-Computer Studies, 43(5–6):907–928, 1995. special issue on the role of formal ontology in the information technology.

• [G21] D. Gabbay. What is negation in a system 2020. Journal of Applied Logic, 8(7):1977–2034, 2021.

• [Q53] W. van Orman Quine. Logic and the reification of universals. In From a Logical Point of View. : 9 Logic-philosophical Essays, 1953.

• [MK01] J. Ma and B. Knight. Reified temporal logics: An overview. Artif. Intell. Rev., 15(3):189–217, May 2001.

• [W3C] https://www.w3.org/TR/swbp-n-aryRelations/

Contact

Mena Leemhuis, M.Sc.
Institut für Informationssysteme
Universität zu Lübeck
Telefon: +49 451 3101 5717
E-Mail: leemhuis(at)ifis.uni-luebck.de
Website: https://www.ifis.uni-luebeck.de/index.php?id=leemhuis

PD Dr. habil.  Özgür Lütfü Özçep
Institut für Informationssysteme
Universität zu Lübeck
Telefon: +49 451 3101 5710
E-Mail: oezcep(at)ifis.uni-luebeck.de
Website: https://www.ifis.uni-luebeck.de/index.php?id=oezcep&L=0

Prof. Dr. Diedrich Wolter
Information Systems and Applied Computer Sciences
Professorship of Smart Environments
Universität Bamberg
Telefon: ++49 (0)951 863 2897
E-Mail: diedrich.wolter(at)uni-bamberg.de
Website: https://www.uni-bamberg.de/en/sme/team/diedrich-wolter/