American University
Browse

Darwinian gradient descent

Download (1.34 MB)
thesis
posted on 2023-08-04, 15:44 authored by Michael Levin

Large optimization problems of many variables can be difficult to solve and very computationally intensive. To dedicate greater computer resources to the problem, this thesis proposes a way to distribute the problem over many different computers using the Berkeley Open Infrastructure for Network Computing (BOINC), an open-source platform where people can volunteer their personal computer to work on various problems while their computer is idle. My program runs a genetic version of a gradient descent algorithm, including conjugate gradient methods, that runs the algorithm in parallel on many computers at once to find a solution faster and to avoid some common problems of gradient descent, such as getting trapped in local minima.

History

Publisher

ProQuest

Language

English

Notes

Thesis (M.S.)--American University, 2010.

Handle

http://hdl.handle.net/1961/thesesdissertations:2855

Media type

application/pdf

Access statement

Unprocessed

Usage metrics

    Theses and Dissertations

    Categories

    No categories selected

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC