Talk:Natural evolution strategy

Computing: Software / CompSci Low‑importance

This article is within the scope of WikiProject Computing, a collaborative effort to improve the coverage of computers, computing, and information technology on Wikipedia. If you would like to participate, please visit the project page, where you can join the discussion and see a list of open tasks.ComputingWikipedia:WikiProject ComputingTemplate:WikiProject ComputingComputing articles

Low

This article has been rated as Low-importance on the project's importance scale.

This article is supported by WikiProject Software (assessed as Low-importance).

This article is supported by WikiProject Computer science (assessed as Low-importance).

Things you can help WikiProject Computer science with:

Here are some tasks awaiting attention:

Article requests :
- Requested articles/Applied arts and sciences/Computer science, computing, and Internet
Cleanup :
- Computer science articles needing attention
- Computer science articles needing expert attention
Copyedit :
- Computing
Expand :
- Computer science
Infobox :
- Computer science articles without infoboxes
Maintain :
- Timeline of computing 2020–present
Photo :
- Find pictures for the biographies of computer scientists (see List of computer scientists)
- Computing articles needing images
Stubs :
- Computer science stubs
Unreferenced :
- WikiProject Computer science/Unreferenced BLPs
Project-related :
- Tag all relevant articles in Category:Computer science and sub-categories with {{WikiProject Computer science}}

An update on using the inverse Fisher information Matrix[edit]

I recently read the paper listed below (information is provided in bibtex form). Specifically the problem reduces significantly when one is using a class of exponential distributions. It says one alternative to premultiplying by the inverse of the Fisher information matrix is to use an update rule in "natural coordinates." The natural coordinates set the Fisher information matrix to the identity matrix. Basically natural coordinates have zero mean and unit variance. They are introduced symbolically, allowing one to take the derivative with respect to them as they vanish. The reparameterized problem looks like
$\mu \leftarrow \mu +A\,\delta ,$
and
$A\leftarrow A\,\exp(M),$
where $\Sigma =A\,A^{\intercal }$ . Then one obtains the update rules by taking the symbolic derivative with respect to $\delta$ and $M$ about zero and the zero matrix. Thus, if my understanding is correct, this results in the Fisher information matrix equaling the identity matrix.

@INPROCEEDINGS{Glasmachers2010,

 author = {Glasmachers, Tobias and Schaul, Tom and Sun, Yi and Wierstra, Daan and Schmidhuber, J\"{u}rgen},
 title = {Exponential natural evolution strategies},
 booktitle = {Proceedings of the 12th annual conference on Genetic and evolutionary computation},
 year = {2010},
 series = {GECCO '10},
 pages = {393--400},
 address = {New York, NY, USA},
 publisher = {ACM},
 acmid = {1830557},
 doi = {10.1145/1830483.1830557},
 isbn = {978-1-4503-0072-8},
 keywords = {black box optimization, evolution strategies, natural gradient, unconstrained optimization},
 location = {Portland, Oregon, USA},
 numpages = {8},
 owner = {Michael},
 timestamp = {2012.02.16},
 url = {http://doi.acm.org/10.1145/1830483.1830557}

}

Natural gradient definition[edit]

The article relies heavily on "natural gradient", however it is not described anywhere in Wikipedia. The linked page doesn't mention it at all. — Preceding unsigned comment added by 193.156.79.200 (talk) 14:15, 10 February 2016 (UTC)[reply]