This article discusses incorporating prior knowledge into data-driven modeling to improve performance, reliability, and generalization. It focuses on the use of manifolds, particularly Lie Manifolds, for modeling finite dimensions and the application of Canonical Correlation Analysis (CCA) in these spaces. The study presents a method to generalize CCA to Lie Manifolds, which improves the ability to make structure-consistent predictions about changes in the state of a robotic hand. The article also explores the Information Bottleneck model and its relevance in dimensionality reduction and information compression methods.
Publication date: 20 Nov 2023
Project Page: Not provided
Paper: https://arxiv.org/pdf/2311.10327
