Our team is formed by computer scientists from Goethe University Frankfurt, and we are in constant dialogue with many collaborators from different domains.
Prof. Gemma Roig
Dr. Karsten Tolle
Mr. Dennis Vetter
Prof. Dr. Roberto Zicari
We are collaborating in a plethora of projects including the following EU funded ones:
eXplainable Artificial Intelligence in Healthcare Management
An Interdisciplinary Master’s Program at the Intersection of AI and Health Care.
Pan-European Response to the Impacts of COVID-19 and future Pandemics and Epidemics (PERISCOPE)
Investigating the broad socio-economic and behavioural impacts of the COVID-19 pandemic, to make Europe more resilient and prepared for future large-scale risks.
We also offer student projects at bachelor and master level at Goethe University in our mission to create awareness of the importance of Trustworthy AI.
Zicari, R. V., Amann, J., Bruneault, F., Coffee, M., Düdder, B., Hickman, E., Gallucci, A., Gilbert, T. K., Hagendorff, T., van Halem, I., Hildt, E., Holm, S., Kararigas, G., Kringen, P., Madai, V. I., Mathez, E. W., Tithi, J. J., Vetter, D., Westerlund, M., & Wurth, R., on behalf of the Z-Inspection® Initiative. (2022). How to Assess Trustworthy AI in Practice (arXiv:2206.09887). arXiv. https://doi.org/10.48550/arXiv.2206.09887
Vetter, D., Tithi, J. J., Westerlund, M., Zicari, R. V., & Roig, G. (2022). Using Sentence Embeddings and Semantic Similarity for Seeking Consensus when Assessing Trustworthy AI. Proceedings of the Workshop on Imagining the AI Landscape after the AI Act (IAIL 2022) Co-Located with 1st International Conference on Hybrid Human-Artificial Intelligence (HHAI 2022). https://ceur-ws.org/Vol-3221/IAIL_paper1.pdf. (Also available on arXiv)
Allahabadi, H., Amann, J., Balot, I., Beretta, A., Binkley, C., Bozenhard, J., Bruneault, F., Brusseau, J., Candemir, S., Cappellini, L. A., Chakraborty, S., Cherciu, N., Cociancig, C., Coffee, M., Ek, I., Espinosa-Leal, L., Farina, D., Fieux-Castagnet, G., Frauenfelder, T., … Zicari, R. V. (2022). Assessing Trustworthy AI in Times of COVID-19: Deep Learning for Predicting a Multiregional Score Conveying the Degree of Lung Compromise in COVID-19 Patients. IEEE Transactions on Technology and Society, 3(4), 272–289. https://doi.org/10.1109/TTS.2022.3195114
Amann, J., Vetter, D., Blomberg, S. N., Christensen, H. C., Coffee, M., Gerke, S., Gilbert, T. K., Hagendorff, T., Holm, S., Livne, M., Spezzatti, A., Strümke, I., Zicari, R. V., & Madai, V. I., on behalf of the Z-Inspection® Initiative. (2022). To explain or not to explain?—Artificial intelligence explainability in clinical decision support systems. PLOS Digital Health, 1(2), e0000016. https://doi.org/10.1371/journal.pdig.0000016
Zicari, R. V., Ahmed, S., Amann, J., Braun, S. A., Brodersen, J., Bruneault, F., Brusseau, J., Campano, E., Coffee, M., Dengel, A., Düdder, B., Gallucci, A., Gilbert, T. K., Gottfrois, P., Goffi, E., Haase, C. B., Hagendorff, T., Hickman, E., Hildt, E., … Wurth, R. (2021). Co-Design of a Trustworthy AI System in Healthcare: Deep Learning Based Skin Lesion Classifier. Frontiers in Human Dynamics, 3, 40. https://doi.org/10.3389/fhumd.2021.688152
Zicari, R. V., Brusseau, J., Blomberg, S. N., Christensen, H. C., Coffee, M., Ganapini, M. B., Gerke, S., Gilbert, T. K., Hickman, E., Hildt, E., Holm, S., Kühne, U., Madai, V. I., Osika, W., Spezzatti, A., Schnebel, E., Tithi, J. J., Vetter, D., Westerlund, M., … Kararigas, G. (2021). On Assessing Trustworthy AI in Healthcare. Machine Learning as a Supportive Tool to Recognize Cardiac Arrest in Emergency Calls. Frontiers in Human Dynamics, 3, 30. https://doi.org/10.3389/fhumd.2021.673104
Zicari, R. V., Brodersen, J., Brusseau, J., Düdder, B., Eichhorn, T., Ivanov, T., Kararigas, G., Kringen, P., McCullough, M., Möslein, F., Mushtaq, N., Roig, G., Stürtz, N., Tolle, K., Tithi, J. J., van Halem, I., & Westerlund, M. (2021). Z-Inspection®: A Process to Assess Trustworthy AI. IEEE Transactions on Technology and Society, 2(2), 83–97. https://doi.org/10.1109/TTS.2021.3066209
* Ethical Implication of AI: Assessing Trustworthy AI in Practice
— Series of Lectures —
This course introduces students to the ethical foundations of Trustworthy AI, and teaches how to assess Trustworthy AI systems in practice by using the Z-Inspection® process.
Course materials and recordings from past lectures are openly available:
Winter 2022/23 @ Seoul National University
Summer 2022 @ Seoul National University
Summer 2020 @ Goethe University Frankfurt
* Trustworthy AI for Healthcare Management
— Online Course —
This MOOC was created as part of the EU project PERISCOPE. Itgives an introduction to trustworthy artificial intelligence and its application in healthcare. This includes modules on basics of artificial intelligence and an introduction to trustworthy and ethical applications of artificial intelligence. A dedicated lesson will present the Z-Inspection® process for assessing trustworthy AI, and real-world case studies will illustrate how to apply the knowledge.
Course available for free on coursera.org.
* Certified Z-Inspection® Teaching Experts:
- Karsten Tolle, Goethe University Frankfurt, Germany
- Dennis Vetter, Goethe University Frankfurt, Germany
- Gemma Roig, Goethe University Frankfurt, Germany
We frequently have projects available for students interested in working with Trustworthy AI. Send us an email to get in touch!
* Systematic Evaluation of Graph Sampling Methods
Bachelor Thesis in co-operation with the Research Group for Theoretical Computer Science
In this work, we comprehensively evaluate the node2vec and the CrossWalk random walk-based sampling for generating graph embeddings of social networks. The resulting embeddings were systematically examined concerning their fairness and accuracy. We will shown that the configuration of the hyperparameters of node2vec and CrossWalk significantly affects the resulting graph representations in both directions and thus either increases or decreases the prediction accuracy or the fairness for selected features.
* The influence of Deep Neural Network architectures on the classification Fairness of face recognition tasks
In this work we will look at Deep Learning tasks with the focus on face recognition and facial attribute analysis. Models that use biased datasets lead to inherit this bias resulting in unfair performance. The goal of this work is to recognize unfair machine learning systems by defining different Fairness measures and evaluating current architectures, to find ways how these architectures can be adapted to perform with better fairness.
* Subjective evaluation of AI explainability methods and their applicability to chest x-rays
Research Project in co-operation with the Frankfurt Big Data Lab
The work presented in this report spans the implementation of a pre-trained deep neural network for image classification, the comparative evaluation of different explainability approaches as well as the integration and evaluation of a metric for quantitatively assessing the quality of provided explanations. Finally, the evaluation of explanation techniques is also performed with a real-world AI system.
* The influence of dataset biases on classification fairness
This work will explore the influence of dataset biases on machine learning outcomes, by selectively subsampling the datasets, and therefore, by artificially introducing biases to the models which are then trained on these datasets. Hence, this thesis is looking for an algorithmic solution to classification fairness. The effects will then be analyzed on different model types.
* Do Machine Learning classifiers have an innate fairness?
In this work, I will analyze different machine Learning classifiers in regard to
their fairness. I will come to the conclusion, that they indeed have innate
fairness and will give recommendations on what classifier to use to satisfy which fairness constraint.
* Ethical assessment of AI systems in healthcare: A use case
The aim of this work is to examine an AI system in the healthcare sector for supporting processes in the detection of OHCA (Out-of-Hospital Cardiac Arrest) with regard to ethical issues. Since the present use case was developed and tested in Europe and thus European values and norms should be respected, the focus is placed on ethical principles of European guidelines for an ethically compliant AI system and specifically on the principles of fairness and explicability for a trustworthy AI system.
We are always eager to collaborate with new people, work on new use-cases or hear about your feedback. Send us an email to get in touch!
Site was created with Mobirise