Oscar Carlsson, a doctoral researcher at the University of Gothenburg, has developed a mathematical framework to improve how neural networks handle geometric transformations and symmetries. His work bridges theory and practical applications, addressing challenges such as rotation invariance in image recognition and curvatures in spherical data. The framework also guides the construction of nonlinear layers with group symmetries, offering potential improvements for advanced classification systems. Carlsson defended his dissertation, Geometry and Symmetry in Deep Learning: From Mathematical Foundations to Vision Applications, on August 29, 2025, and will continue teaching mathematical analysis to mechatronics students.
Source: University of Gothenburg
About the Author
