[Illinois]: Perturbative Reinforcement Learning Using Directed Drift

By AbderRahman N Sobh

University of Illinois at Urbana-Champaign

This tool trains two-layered networks of sigmoidal units to associate patterns using a real-valued adaptation of the directed drift algorithm.

Launch Tool

This tool version is unpublished and cannot be run. If you would like to have this version staged, you can put a request through HUB Support.

Archive Version 1.0c
Published on 19 Aug 2013 All versions

doi:10.4231/D3MP4VN4X cite this

This tool is closed source.

Category

Tools

Published on

Abstract

From Tutorial on Neural Systems Modeling, Chapter 7: In the directed drift algorithm (Venkatesh 1993), input patterns are presented to the network, and one or sev­eral randomly chosen weights have their binary values flipped if the output is in error, but the weights are left unperturbed otherwise. Directed drift is proven to work in this restricted context (Venkatesh 1993). We explore its use for real-valued weights in the next example.

Sponsored by

NanoBio Node, University of Illinois Champaign-Urbana

Cite this work

Researchers should cite this work as follows:

  • Tutorial on Neural Systems Modeling, Copyright 2010 Sinauer Associates Inc. Author: Thomas J. Anastasio

  • AbderRahman N Sobh (2014), "[Illinois]: Perturbative Reinforcement Learning Using Directed Drift," https://nanohub.org/resources/pertdd. (DOI: 10.4231/D3MP4VN4X).

    BibTex | EndNote

Tags

  1. Illinois
  2. MCB 493
  3. nano/bio
  4. NanoBio Node at Illinois
  5. Neural Systems Modeling