Christoph Wehner

How to incorporate human knowledge and preferences in machine learning methods?

Main Research Interests

• Interactive and Interpretable Machine Learning on Knowledge Graphs

Methods

• Knowledge Graphs
• Link Prediction
• Attribution Methods
• Reinforcement Learning
• Game Theory
• Preference-Based Learning

Advised Student Theses

• Text Literal Supported Link Prediction in Knowledge Graphs Utilizing the MINERVA Algorithm by Julian Holger Fehr
• Evaluating the Faithfulness of XAI Attribution Methods -- End-To-End Testing Pipeline of Relevance-Based Attribution Methods on Computer Vision Foundation Models by Bartu Soykök

Open Topics

Interactive Root Cause Analysis for the E-Drive Manufacturing Line @ BMW Group 

For more infos click here.

Rule-Guided Reinforcement Learning for Link Prediction -- Extending MINERVA to Human-in-the-Loop Reinforcement Learning:

The thesis/project aims to combine MINERVA [1] with a rule-based human-in-the-loop reinforcement learning approach [2].
MINERVA [1] is a reinforcement learning approach that suggests missing relations between two entities of a knowledge graph (also known as link prediction / knowledge graph completion).
The rule-based human-in-the-loop reinforcement learning approach from [2] shows how knowledge from an expert can be tapped in the form of examples. The examples are then generalized to rules and fed into the learning process of a reinforcement learner. With the expert knowledge, the agent can reach an optimal learning result faster.
The student will first implement MINERVA [1] on the benchmark knowledge graph like DBPedia50.
She will then explore how the approach from [2] can be applied to MINERVA. Most ideas will be transferable, but a new method of how the rules can influence the probability of the action chosen by the reinforcement agent is needed.
The scientific added value of the work can be found in the fact that the student will develop the first human-in-the-loop link prediction approach with rule-based feedback.

[1] Das, R., Dhuliawala, S., Zaheer, M., Vilnis, L., Durugkar, I., Krishnamurthy, A., Smola, A., McCallum, A. (2018). Go for a walk and arrive at the answer: Reasoning over knowledge bases with reinforcement learning. In ICLR 2018.

[2] Bignold, A., Cruz, F., Dazeley, R., Vamplew P., Foale C. (2021). Persistent rule-based interactive reinforcement learning. In Neural Computing & Applications 2021.

Are you interested in writing your thesis or Project on one of the latter topics? And you want to solve real-world problems regarding root cause analysis in the manufacturing process of electric vehicles?
Then feel free to contact me!

Peer Reviewed

• Wehner, C., Powlesland, F., Altakrouri, B., Schmid, U. (2022). Explainable Online Lane Change Predictions on a Digital Twin with a Layer Normalized LSTM and Layer-wise Relevance Propagation. In: Fujita, H., Fournier-Viger, P., Ali, M., Wang, Y. (eds) Advances and Trends in Artificial Intelligence. Theory and Practices in Artificial Intelligence. IEA/AIE 2022. Lecture Notes in Computer Science, vol 13343. Springer, Cham. https://doi.org/10.1007/978-3-031-08530-7_52

Under Review

• Eirich, J., Jäckle, D., Sedlmair, M., Wehner, C., Schmid, U., Bernard, J., Schreck, T. (2022). ManKnowVis: How to Support Different User Groups in Contextualizing and Leveraging Knowledge Repositories. – under review IEEE TVCG

• Schramm, S., Wehner, C., Schmid, U. (2022). Comprehensible Artificial Intelligence on Knowledge Graphs: A Survey. – under review Journal of Web Semantics

• Wehner, C., Kertel, M., Wewerka, J. (2022). Interactive and Intelligent Root Cause Analysis in Manufacturing with Causal Bayesian Networks and Knowledge Graphs. – under review IEEE VTC 2023 Spring

Posters & Short Papers

• Wehner, C. (2022). Interactive and Explainable Link Prediction in Knowledge Graphs. Pattern Recognition and Artificial Intelligence. ICPRAI 2022. Doctoral Consortium.

• Dialogtag des BMW Doktoranden-Programms in Munich, Germany, September 2022
• HCAI-LAB Symposium “Digital transformation for smart farm and forest operations” in Tulln an der Donau, Greater Vienna Area, Austria, August 2022
• DeepLearn Summer School 2022 in Las Palmas de Gran Canaria, Spain, July 2022
• ICPRAI2022 Doctoral Consortium in Paris, France, May 2022