Arsen Sheverdin

About

I am a Doctoral Researcher at the Helmholtz Munich and Technical University of Munich (TUM) at Dr. Vincent Fortuin's lab .

My research interests are in LLM trustworthiness, uncertainty estimation, and LLM efficiency. Currently, I study how LLMs can express uncertainty reliably in natural language, and methods for ensuring calibrated and verifiable uncertainty communication. I am also interested in probabilistic approaches to test-time inference for decision-making and efficient deployment. More broadly, I work on generative modeling and efficient LLMs.

Bio

Previously I graduated as an ELLIS Honours MSc in Artificial Intelligence at the University of Amsterdam, where I wrote my thesis on probabilistic deep learning methods for continual learning with Dr. Eric Nalisnick at the AMLab . I obtained my undergraduate degree in Physics with Mathematics minor at the Nazarbayev University. During my undergraduate, I was fortunate enough to receive the Yessenov Foundation scholarship to conduct a research visit at the Cornell University, where I worked on developing deep learning methods for inverse design for nanophotonic structres, under the supervision of Dr. Francesco Monticone . As an undergraduate, for two years I also was a Research Assistant at my university's Physics department, where I studied the nanophotonic structures and their optimization under the supervision of Dr. Constantinos Valagiannopoulos