The rotation-symmetric spherical cow of Deep Learning $\rightarrow$ Wide Learning

Abstract: Due to the current boom in AI, deep learning has become mainstream. The name is inspired by the fact that the corresponding machine learning models are usually a composition of many layers of parametrized transformations. To swim against (or arguably orthogonal to) the tide, what happens if we instead consider particularly 𝘸𝘪𝘥𝘦 layers? In this talk, I will present the spherical cow equivalent of neural networks and how this spherical cow can even be endowed with additional symmetry, like 3d rotations (... the analogy may be lacking).

Mathematics

Audience: general audience

( paper )

Gothenburg PhD seminar

Series comments: Rooms and times may vary, please check the latest update. In-person only.