saba-sadiya [at] em.uni-frankfurt.de
Download CV
The ARSU AI Lab Blog
My band: Beirut 66!
Arabic, Hebrew, English, German
Robert-Mayer-Straße 12
60325 Frankfurt am Main, DE
Tweets by SariSabaSadiya
About
I am a Principle Investigator at Constructor Knowledge Labs focusing on Human-AI Fusion & Cognitive Wearable Intelligence. Previously, I was a Research fellow working at Dr Gemma Roig's Computational Vision & AI lab at the Frankfurt Goethe University, and Dr Radoslaw Cichy's Neural Dynamics of Visual Cognition lab at Berlin.
I originally studied Mathematics and Computer Science, followed by a 2 years stint at Apple. I left industry to become a Fulbright grantee at Michigan, where I completed a dual Ph.D in Computer Science under Dr Mohammad Ghassemi, and Cognitive Neuroscience under Dr Taosheng Liu. My PhD thesis focused on utilizing electroencephalography in conjunction with state-of-the-art machine learning techniques to investigate how attention modulates sensory representations.
My Algorithms for Reasoning, Sensing, and Understanding (ARSU) lab focuses on interpretability for understanding emergent representations, and computational approaches in bioinformatics.
I spend my free time learning to play various instruments to varying degrees of success.
Project Highlights
Efficient Shortcut and Bias Mitigation
Even SOTA models like CLIP are riddled with shortcuts (In ImageNet, a watermark in Hanzi correlates with the 'cardboard' class). I develop methods to mitigate such biases, focusing on interpretability, label efficiency, and resources accessibility.
Papers:Efficient unsupervised shortcut learning detection and mitigation in transformers In Proceedings of the 2025 International Conference on Computer Vision (ICCV), 2025 L. Kuhn*, S. Saba-Sadiya*, J. Schlotterer, F. Buettner, C. Seifert, R. Gemma
[pdf] [bib] [code]Weakly Supervised Shortcut Mitigation Using Sparse Projections Online Preprint, 2025 M. Ahsan, D. Towadras, S. Saba-Sadiya, J. Schlotterer, C. Seifert, R. Gemma
[pdf] [bib] [code]
Feature Imitating Networks
Novel framework for integrating expert knowledge into deep learning networks by initalizing weights to approximate known statistical measures.
Papers:Feature Imitating Networks In Proceedings of the 2022 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2022 S. Saba-Sadiya and T. Alhanai and T. Liu and M. Ghassemi
[pdf] [bib] [code]MambaNet: A Hybrid Neural Network for Predicting the NBA Playoffs In SN Computer Science 5 (5) , 2022 R. Khanmohammadi, S. Saba-Sadiya and T. Alhanai and T. Liu and M. Ghassemi
[pdf] [bib] [code]
Artifact Detection and Correction in EEG Data
Applying state-of-the-art self-supervised machine learning techniques to artifact detection and correction in EEG data.
Papers:Different Algorithms (Might) Uncover Different Patterns: A Brain-Age Prediction Case Study In Proceedings of the IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 2023 T. Ettling*, S. Saba-Sadiya*, and G. Roig
[pdf] [bib] [code]Unsupervised EEG Artifact Detection and Correction In Frontiers in Digital Health, Special issue on Machine Learning in Clinical Decision-Making. Volume 2, 2020 S. Saba-Sadiya and E. Chantland and T. Alhanai and T. Liu and M. Ghassemi
[pdf] [bib] [code]EEG Channel Interpolation Using Deep Encoder-Decoder Networks In Proceedings of the 2020 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), 2020 S. Saba-Sadiya and T. Alhanai and T. Liu and M. Ghassemi
[pdf] [bib] [code]
