REES: Reasoning Engines Evaluation Shell

1
REES Reasoning Engines Evaluation Shell

• version 3.0

Automated Reasoning Lab University of
California, Irvine
2
Credits

• Team Members
• Radu Marinescu
• Kalev Kask (PI)
• Rina Dechter (PI)
• Kyle Bollen
• Plugins by
• Kalev Kask
• Radu Marinescu
• Dan Frost

3
Introduction

• Knowledge Representation Framework (Graphical
  Models)
• Constraint Networks (CSP)
• Constraint Optimization (COP)
• Belief Networks (BN)
• Queries
• CSP, Max-CSP
• Belief Updating (BEL)
• Most Probable Scenario (MPE, MAP)

Within this framework a reasoning engine
represents a collection of algorithms that
implements a specific query type.
4
Motivation

• Software environment to support research and
  development in the area of automated reasoning.
• Plug-in architecture
• Rapid integration of different libraries of
  reasoning algorithms (plug-ins).
• Common workbench that promotes joint evaluation
  and comparison of alternative technologies.

5
Architecture - Overview
Plugin 1
Plugin 1
Plugins C/C implementations of reasoning
algorithms
Plugin 1

Plugin 1
Microsoft Windows implementation
6
REES Suite 3.0
7
REES 3.0 System Features

• User friendly interface
• REES Assistants (wizards)
• Model setup
• Parametric (configure internal generators)
• Complete (use the network editor)
• Task setup
• Choose/configure processing algorithms
• REES Network Editor
• Bayesian networks
• Constraint networks
• Flexible Plug-in Architecture
• Multiple plugins sharing the same interface and
  data

8
REES Network Editor

• Belief Networks
• Graph structure toolbox
• Node/CPT properties editor
• Constraint Networks
• Graph structure toolbox
• Node/Constraint properties editor

9
REES Assistants (Wizards)

• Model Setup Wizard
• Assists the configuration of a random problem
  generator
• Graphical structure
• Functional structure
• Task Setup Wizard
• Assists the configuration of an experiment
• List of algorithms
• Input/Output params

10
Task Execution and Results

• Task Execution
• REES maintains a detailed log of the execution.
• Multithreaded execution allows immediate
  interruption.
• Task Results
• Spreadsheet-like visualization of results (each
  column represents an output parameter).
• Simple and intuitive comparison between
  algorithms.

11
Summary

• Plug-in architecture allows multiple reasoning
  engines to share the same interface and data.
• Joint algorithmic evaluation and comparison now
  possible with REES Suite 3.0
• Public release is available at
  http//www.ics.uci.edu/radum/rees.html