ML discussion circle

​This is a weekly virtual study group dedicated to the discussion of recent landmark papers at the intersection of Mathematics, Machine Learning, Physics and Chemistry. Objective is to have a interdisciplinary discussion on the mathematical foundations, statistical rigor, and physical constraints of state of the art deep learning architectures. We prioritize theoretical discussion and exploring ideas over software principles.

​The First Paper

​We will begin by dismantling: "Geometric Algebra Transformer" (GATr) (Brehmer et al., 2023)"

​Most neural networks treat 3D data as simple vectors in real space. GATr instead represents data as multivectors in a Projective Geometric Algebra. We will discuss how this shift allows for a unified representation of points, lines, and planes, and how it naturally enforces E(3) equivariance.

​Participant Profile

• ​Mathematical Background: Proficiency in Linear Algebra and foundational Deep Learning is required.

• ​Domain Interest: Ideal for those with background/interest in Mathematics, Computer Science, Physics, Chemistry.

• ​No Code Required: This is a "math first" group. You do not need to be a programmer to participate.

• ​A genuine curiosity and a willingness to engage deeply with the mathematical beauty of the subject are the only true requirements for joining us.