http:^^envy.cs.umass.edu^

来自「This data set contains WWW-pages collect」· EDU^ 代码 · 共 114 行

Date: Wed, 20 Nov 1996 20:01:20 GMT
Server: NCSA/1.5.2
Last-modified: Tue, 01 Oct 1996 17:35:33 GMT
Content-type: text/html
Content-length: 4272

<TITLE>Adaptive Networks Lab Home Page</TITLE>

<!WA0><IMG SRC="http://envy.cs.umass.edu/anwlogo2.gif">
<!--
  -- Commented out old gif and replaced with logo.  8/29/95.  mcnulty.
  -- <!WA1><IMG SRC="http://envy.cs.umass.edu/anw-home-page-icon.gif"><H1>Adaptive Networks Lab</H1><p>
  -- 
<hr>

<!WA2><IMG SRC="http://envy.cs.umass.edu/public_html/images/construction.gif" WIDTH=40 HEIGHT=40 ALIGN=middle>Warning: this page is still under construction!
  -->


<hr>

Welcome to the ANW Laboratory at UMASS/Amherst co-directed by 
<!WA3><A HREF="http://envy.cs.umass.edu/People/barto/barto.html">Prof. Andrew G. Barto</A> and
<!WA4><A HREF="http://envy.cs.umass.edu/People/sutton/sutton.html">Richard S.
Sutton</A>

<UL>
<LI> <!WA5><A HREF="http://envy.cs.umass.edu/anw-home-page.html#Lab Introduction">Introduction</A>
<LI> <!WA6><A HREF="http://envy.cs.umass.edu/People/people.html">People</A>
<LI> <!WA7><A HREF="http://envy.cs.umass.edu/Publications/publications.html">Publications</A>
<LI> <!WA8><A HREF="http://envy.cs.umass.edu/anw-home-page.html#Current Projects">Current Projects</A>
<LI> <!WA9><A HREF="http://envy.cs.umass.edu/Places/places.html">Related Web Sites</A>
<LI> <!WA10><A HREF="http://envy.cs.umass.edu/restrict/anw.faq/">Information For ANW users</A> (restricted to ANW users)
</UL>

<hr>

<A NAME="Lab Introduction">
<b>Introduction to the Adaptive NetWork Laboratory</b>
</A><p>

The major focus of the Adaptive NetWork (ANW) Laboratory is the
learning paradigm known in artificial intelligence as <!WA11><a
href="http://envy.cs.umass.edu/reinforcement-learning.html">reinforcement learning</a> (RL).
The group's research began in 1977 with a reassessment of an approach
to artificial intelligence using artificial neural networks, and its
first publications appeared in 1981.  Laboratory researchers
contributed to the re-emergence of this approach in artificial
intelligence, with an emphasis on methods for network learning through
realistic interaction with dynamic environments and without the help
of knowledgeable teachers.

<hr>

<A NAME="Current Projects"><b>Current Projects</b></A>

<UL>
<LI>  <b>Basic research in the theory and applicaton of RL algorithms based on
Dynamic Programming.</b><p> The objectives of this project are: 
<OL>
<LI> to continue development of DP-based RL methods and their theory, 
<LI> to explore their utility in distributed control architectures, and
<LI> to obtain better characterizations of the problems for which they
 are best suited. 
</OL>
<p>

<LI> <b>Multiple time scale reinforcement learning.</b><p> 
This project investigates a new approach to
learning models of dynamical systems for use in advanced RL architectures.
We are developing a method by which an RL system can learn multiple time scale models and use them as
the basis for hierarchical learning and planning. The project's objectives are to develop the
mathematical theory of this approach, to examine its relationship to control
theory and to behavioral and neural models of learning, and to demonstrate its effectiveness
in a number of simulated learning tasks. 
<p>

<LI> <b>A control basis for learning and skill acquisition.</b>
<p>

This project is being conducted in collaboration with <!WA12><A
HREF="http://piglet.cs.umass.edu:4321/~grupen/home.html">Professor R.
Grupen</A> of the <!WA13><A HREF="http://piglet.cs.umass.edu:4321/lpr.html">
Laboratory for Percepual Robotics</A>. This project addresses
principles for organizing purposeful, coordinated action for complex
sensorimotor systems operating in unstructured environments. It
emphasizes adaptive skill acquisition through the activation of
combinations of reuseable feedback control laws.  <p>

<LI> <b>Modeling cerebellar and premotor circuits.</b><p>

This project is being conducted as part of the Center for Neuroscience
Research on Neuronal Populations and Behavior, directed by James C.
Houk, Professor and Chair of Physiology and Professor of Biomedical
Engineering, Northwestern University Medical School, Chicago, IL. The
overall objective of this project is use mathematical and computer
models to refine and test hypotheses about how the cerebellum and
motor cortex function together to support motor activities. The key
hypothesis is that these networks can be modeled as arrays of
adjustable pattern generators capable of storage, recall and execution
of motor programs.  Understanding how motor programs can be learned
has also been a focus of this study.

<hr>
<!WA14><a href="http://www.cs.umass.edu">Back</a> to the CS home page
<hr>
Comments? Please contact Nathan Sitkoff
(<em>[e-mail]</em>)<p>

<i>Last Update: 1/11/96</i>