Kosio Beshkov

Kosio Beshkov is a Postdoctoral Fellow in Condensed Matter Physics at the University of Oslo, specializing in the intersection of topological data analysis and machine learning. His research focuses on theoretical frameworks for understanding neural network representations and biological neural systems.

His primary research interests include:

• Theory of deep neural networks in overparametrized regimes
• Topological data analysis of neural manifolds
• De novo protein design using evolutionary algorithms and geometric modeling
• Connections between network representations and topological spaces

Recent publications demonstrate strong trends in computational neuroscience, with 7 papers from 2021-2025 spanning journals like PLoS Computational Biology and iScience. His work consistently applies polyhedral geometry, quotient spaces, and homology to neural representation problems, while expanding into protein language models and gene therapy applications.

Current technical approaches combine:

• Topological data analysis for high-dimensional neural data
• Geometric deep learning for robust representations
• Biophysically-detailed neuron modeling
• Protein structure-geometry relationships