Sean a bittle mark s fox march 7th 2009
Download
1 / 11

Sean A. Bittle Mark S. Fox March 7th, 2009 - PowerPoint PPT Presentation


  • 51 Views
  • Uploaded on
  • Presentation posted in: General

Introducing Constrained Heuristic Search to the Soar Cognitive Architecture (demonstrating domain independent learning in Soar) The Second Conference on Artificial General Intelligence, AGI-09. Sean A. Bittle Mark S. Fox March 7th, 2009. 1 /11. The Problem.

loader
I am the owner, or an agent authorized to act on behalf of the owner, of the copyrighted work described.
capcha

Download Presentation

Sean A. Bittle Mark S. Fox March 7th, 2009

An Image/Link below is provided (as is) to download presentation

Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author.While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server.


- - - - - - - - - - - - - - - - - - - - - - - - - - E N D - - - - - - - - - - - - - - - - - - - - - - - - - -

Presentation Transcript


Introducing Constrained Heuristic Search to the Soar Cognitive Architecture(demonstrating domain independent learning in Soar) The Second Conference on Artificial General Intelligence, AGI-09

Sean A. Bittle

Mark S. Fox

March 7th, 2009

1/11


The Problem

  • General problem solving and learning are central goals of AI research on cognitive architectures

  • However, there are few examples of domain independent learning in cognitive architectures

The Goal

  • Demonstrate Soar can learn and apply domain independent knowledge

    But to achieve this goal we need to augment the Soar’s problem-solving paradigm (CHS-Soar)

2/11


Soar Cognitive Architecture

  • Developed by Newell, Laird and Rosenbloom at CMU, 1983

  • Symbolic Cognitive Architecture where all long term knowledge is encoded as productions rules.

  • Based on the hypothesis that all goal-oriented behavior can be cast as the selection and application of operators to a state in a problem space

3/11


Constraint Graph

Va

Ci

Vc

Cii

Vb

Constrained Heuristic Search (CHS)

  • Developed by Fox, Sadeh and Bayken, 1989

  • CHS is a problem solving approach that combination of constraint satisfaction and heuristic search where the definition of the problem space is refined to include:

    • Problem Topology

    • Problem Textures

    • Problem Objective

  • CP/CHS allows us to employ a generalized problem representation (CSP) and utilize generic, yet powerful problem solving techniques

4/11


CHS-Soar

“What are we trying to learn?”

5/11


CHS-Soar

What are “Texture Measures?”

  • Minimum Remaining Values (MRV) – variable selection

  • Degree (DEG) – variable selection

  • Least Constraining Value (LCV) – value selection

6/11


CHS-Soar

“How Do We Select a “Good” Texture Measure?”

7/11


CHS-Soar

“What Do We Learn...Again?”

Traditional Soar Agent Chunks tend to include domain specific knowledge

Hyper-heuristics:

heuristics to choose heuristic measures

8/11


Experiments

Three experiments conducted to investigate:

  • Integration of rule and constraint based reasoning

  • Domain Independent Learning

  • Scalability of externally learned chunks

    Problem types being considered:

  • Job Shop Scheduling (JSS)

  • Map Colouring

  • Radio Frequency Assignment Problem (RFAP)

  • N-Queens, Sudoku, Latin Square

  • Towers of Hanoi, Water Jugs

  • Configuration Problems

  • Random CSPs

9/11


Results: Domain Independent Learning

Map Colouring(n = 11)

Job Shop Scheduling(n = 15)

RFAP(n = 200)

10/11


Conclusions

  • Demonstrated integration of rule and constraint based reasoning

  • Demonstrated the ability to learn rules while solving one problem type that can be successfully applied in solving another problem type

  • Demonstrated ability to discover, learn and use multi-textured “hyper-heuristics” leading to improved solutions over traditional unary heuristics

11/11


ad
  • Login