Feature Geometry
A mathematical framework for feature-centric information processing, which
- formulates reprentation learning as information decomposition
- separates feature learning and feature usages
- provides principled algorithm designs
- for learning multivariate dependence structures
- with deep neural networks as building blocks
Applications
- sequential dependence decomosition (Allerton 2023)
- multiuser detection in wireless communication (Allerton 2023)
- understanding kernel methods (ISIT 2023)
Full paper: Neural Feature Learning in Function Space. (JMLR, vol 25:142)
More information can be found in this 
