English

A Common Framework for Natural Gradient and Taylor based Optimisation using Manifold Theory

Machine Learning 2018-10-04 v3 Machine Learning

Abstract

This technical report constructs a theoretical framework to relate standard Taylor approximation based optimisation methods with Natural Gradient (NG), a method which is Fisher efficient with probabilistic models. Such a framework will be shown to also provide mathematical justification to combine higher order methods with the method of NG.

Cite

@article{arxiv.1803.09791,
  title  = {A Common Framework for Natural Gradient and Taylor based Optimisation using Manifold Theory},
  author = {Adnan Haider},
  journal= {arXiv preprint arXiv:1803.09791},
  year   = {2018}
}
R2 v1 2026-06-23T01:05:41.459Z