Grigoris is an Assistant Professor at the University of Wisconsin–Madison, where his research centers on reliable and trustworthy machine learning. His work seeks to develop robust models capable of maintaining high performance in the presence of noise and out-of-distribution data.
His research spans two key areas:
- Parsimonious learning: Grigoris explores the inherent low-rank and sparse structures present in natural signals such as text and images. Through empirical and theoretical analysis, he investigates the inductive biases, expressivity, trainability, and generalization properties of neural and polynomial networks.
- Trustworthy models: He focuses on enhancing the robustness and extrapolation capabilities of machine learning models, with particular interest in defending against adversarial attacks. His recent work includes adversarial robustness in the text domain. In the long term, he aims to develop models that are privacy-preserving, robust to malicious inputs, and capable of strong generalization to novel scenarios.
Grigoris’s research contributes to building machine learning systems that are not only accurate, but also reliable, interpretable, and secure.