Dr. Sarah Fabi

Machine learning.

In the concluding stages of my Ph.D. in psychology, I self-taught a range of machine learning techniques, with a particular focus on artificial neural networks. This initiative led me to broaden my research scope by pursuing a postdoctoral position in computational modeling approaches. At the University of Tübingen, I explored enhancements to artificial neural networks through the integration of inductive biases inspired by human cognition.

My research journey included a tenure as a visiting scholar in Prof. Virginia de Sa’s lab at the Halıcıoğlu Data Science Institute at UC San Diego. Here, I leveraged a computer vision model to address a psychological research question, an endeavor supported by a fellowship from the Science department of the University of Tübingen. Additionally, a visit to Stanford University enabled me to examine machine intuition in large language models through psychological methods, contributing to a publication in Nature Computational Science.

These experiences paved the way to my current position as an AI Researcher at Mercedes-Benz AG, where I am tasked with developing the Mercedes Virtual Assistant as well as making generative AI usable for the company.

Peer-reviewed papers:

• Hagendorff, T., Fabi, S., & Kosinski, M. (2023). Machine intuition: Uncovering human-like intuitive decision-making in GPT-3.5. Nature Computational Science.

• Hagendorff, T. & Fabi, S. (2023). Why we need biased AI – How including cognitive biases can enhance AI systems. Journal of Experimental & Theoretical Artificial Intelligence.

• Raina, R., Monares, M., Xu, M., Fabi, S., Xu, X., Li, L., Sumerfield, W., Gan, J., & Virginia R. de Sa. (2022). Exploring Biases in Facial Expression Analysis using Synthetic Faces. In NeurIPS Workshop SyntheticData4ML.

• Fabi, S., Xu, X., & de Sa, V.R. (2022). Exploring the racial bias in pain detection with a computer vision model. Proceedings of the Annual Meeting of the Cognitive Science Society, 44.

• Fabi, S., Holzwarth, L., & Butz, M.V. (2022). Efficient learning through compositionality in a CNN-RNN model consisting of a bottom-up and a top-down pathway. Proceedings of the Annual Meeting of the Cognitive Science Society, 44.

• Fabi, S., Otte, S., Scholz, F., Wührer, J., Karlbauer, M., & Butz, M.V. (2022). Extending the Omniglot Challenge: Imitating handwriting styles on a new sequential data set. IEEE Transactions on Cognitive and Developmental Systems.

• Fabi, S., Otte, S., & Butz, M.V. (2021). Compositionality as learning bias in generative RNNs solves the Omniglot challenge. In International Conference on Learning Representations (ICLR) – Workshop Learning to Learn.

• Fabi, S., Otte, S., & Butz, M.V. (2021). Fostering compositionality in latent, generative encodings to solve the Omniglot challenge. In I. Farkas, P. Masulli, S. Otte, & S. Wermter (Eds.), Proceedings of Artificial Neural Networks and Machine Learning – ICANN 2021, Part II, 525-536.

• Hobbhahn, M., Butz, M.V., Fabi, S., & Otte, S. (2020). Sequence classification using ensembles of recurrent generative expert modules. In Proceedings of the 28th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning – ESANN 2020, 333-338.

• Fabi, S., Otte, S., Wiese, J.G., & Butz, M.V. (2020). Investigating efficient learning and compositionality in generative LSTM networks. In I. Farkas, P. Masulli, & S. Wermter (Eds.), Proceedings of Artificial Neural Networks and Machine Learning – ICANN 2020, 143-154.

Media appearances:

• Murphy, H. T. (10.12.2022) Who Painted That New Cosmic You? – Lensa’s trippy, A.I.-generated avatars are a viral hit. But their magic comes from an unsettling place. Slate.

Presentations:

• Fabi, S. (2022). Applying Cognitive Science to Machine Learning and vice versa. Research Talk, DeepMind, London.

• Fabi, S. (2022). Efficient learning in generative RNNs: Solving one-shot tasks by including compositionality as an inductive bias. Tech Talk, Amazon, Tübingen.

• Fabi, S. (2022). Machine learning for psychological research: Using the example of the racial bias in pain recognition. Machine Learning in Science: Postdoc Symposium of the Cluster of Excellence, Tübingen.

• Fabi, S., Otte, S. & Butz, M.V. (2021). Fostering compositionality in generative RNNs to solve the Omniglot challenge. Oral presentation at the Computational Cognition Workshop.

• Fabi, S., Otte, S., & Butz, M.V. (2021). Fostering compositionality in latent, generative encodings to solve the Omniglot challenge. Oral presentation at the 30th International Conference on Artificial Neural Networks (ICANN).

• Fabi, S., Otte, S. & Butz, M.V. (2021). Does compositionality as a prior in Generative RNNs lead to efficient learning of temporal predictions?. Oral presentation at the ICDL Workshop Spatio-temporal Aspects of Embodied Predictive Processing.

• Fabi,  S.,  Otte,  S.,  & Butz,  M.V. (2021). Compositionality as learning bias in generative RNNs solves the Omniglot challenge. Poster presented at International Conference on Learning Representations (ICLR) – Workshop Learning to Learn.

You can find my Google Scholar profile here.