I am a machine learning researcher with a focus on compression and robust generalization. I am currently a post-doc at Imperial College London (Note: Our paper on compressibility and robustness to spurious correlations have been accepted to ICCV 2025 - link coming soon!). Before this, I was a research scientist intern at DeepMind London, working on robust human-AI complementarity (published in Nature Medicine) in medical systems, as well as robust evaluation and training under uncertain ground truth (published in Medical Image Analysis). I have completed my PhD in Boğaziçi University, with a research focus on robust generalization in deep learning and probabilistic latent variable models.

I have industry experience on research and deployment of machine learning systems, both as a team lead and freelance consultant, involving work on multivariate time series analysis, AutoML system development, and predictive maintenance among others. At the height of my involvement in the industry, I was in charge of the team that oversaw the forecasting of withdrawals from 8000 ATMs daily.

I have a BA and MA in psychology from my alma mater, where I mainly focused on psycholinguistics research, as well as experiment design and statistical inference. My thesis was on embodied language processing, investigating how understanding long texts utilized perceptual representations. In my research visit to UC Berkeley’s Concepts and Cognition Lab, I investigated how different wording choices a language affords can affect a person’s causal explanations.

To learn more, feel free to see my CV, visit my GitHub, Google Scholar, or LinkedIn pages, or e-mail me through melih.barsbey at gmail.com.