Mbwenga Maliti: Convolution on compact groups and natural language processing

Abstract.

There is a mathematical framework that can be used for explicitly constructing neural networks with internal representations of its input that are equivariant to the action of an arbitrarily chosen compact group. I have described parts of it and used a simple model of languages and translations to examine a part of the aforementioned framework necessary for applying it to natural language processing. There is a significant amount of work remaining in order to build a full neural network according to the mathematical framework, as well as in order to incorporate more sophisticated models of language and translations into it.