【报告摘要】Geometric Deep Learning seeks to extend modern machine learning beyond traditional data representations by exploiting the geometric structure underlying the data. Geometric Algebra provides a unified mathematical framework for representing and manipulating geometric objects and transformations, making it a natural candidate for incorporating geometric priors into learning systems. This talk briefly introduces the basic ideas behind neural network architectures based on Geometric Algebra and discusses how geometric structure and symmetry can be encoded directly into their representations and operations. The primary focus is on highlighting recent developments and applications in areas such as computer vision, bioinformatics, and scientific machine learning, illustrating the potential of Geometric Algebra as a foundation for geometrically informed learning.

【报告人简介】Sven Buchholz received his Diploma in Computer Science from the University of Kiel, Germany, and subsequently earned his PhD (summa cum laude) from the same university. For his thesis "A Theory of Neural Computation with Clifford Algebras", he was awarded the Faculty's Best Dissertation Prize in 2005. He went on to hold positions as Senior Researcher at the Free University of Berlin and the University of Magdeburg, both in Germany. In 2009 he was co-founder of a neuro-cognitive research start-up, and served as its CTO till 2014. Since 2014 he is Full Professor of Computer Science at the Technische Hochschule Brandenburg, Germany. He also held visiting appointments as an Academic Visitor at Grenoble Alps University (France), Link?ping University (Sweden) and Technical University Graz (Austria). Professor Buchholz's research centers on Mathematical Foundations of Machine Leaning and Machine Learning for Mathematics. He particularly favors Geometric Algebra for studying these areas.