Clifford Algebras: A Proposal Towards Improved Image Recognition in Machine Learning

摘要

Machine learning algorithms are designed to learn autonomously to learn general rules from a set of examples. The importance of this task lies in its potential to provide future and past predictions, as well as to improve the interpretability of the data. RGB images have shown themselves to be a challenging topic to neural networks as their 3 dimensions (Red, Green and Blue) have to be processed using mathematical techniques designed for 1-dimensional inputs. However, an implementation of neural networks using Clifford algebras can speed up the processing time and improve the performance, as the resulting network is based on a 4-dimensional space.