I am a first-year PhD Student at the Tübingen AI Center, University of Tübingen, supervised by Seong Joon Oh and co-supervised by Matthias Bethge. My personal webpage is at ankitsonthalia.com.

The simplicity bias often causes machine learning models to learn semantically meaningless, “spurious” correlations between inputs and outputs. This phenomenon hurts generalization, thus leading to often unpredictable failure cases and our inability to trust AI systems. I’m interested in understanding how to encourage ML models to learn more meaningful, causal correlations instead. Better generalization capability and trustworthiness would follow as direct consequences.

I obtained my Bachelor and Master degrees in Computer Science and Mathematics, respectively, in 2022, from a five-year integrated program at BITS Pilani, Pilani campus, India. During this time, I worked on a Computer Vision project at the Augmented Vision Department, German Research Centre for Artificial Intelligence (DFKI), led by Prof. Dr. Didier Stricker; and another under the supervision of Bharath Ramesh at the Western Sydney University. After that, I went on to further explore my science and engineering interests as a research software engineer at the Tübingen AI Center, where I collaborated with STAI for a year, before joining the group as a PhD student.