|Muskrat-II from University of Aberdeen|
The MUSKRAT (Multistrategy Knowledge Refinement and Acquisition Toolbox)
framework aims to unify problem solving, knowledge acquisition and machine
learning in a single computational framework (WHITE,
SLEEMAN, 1998). Given a set of Knowledge Bases (KBs) and Problem Solvers
(PSs), the system will try to identify which KBs could be combined with which
PSs to solve a given task.
Towards a Solution
In this research, KBs with certain PSs will be represented as Constraint Satisfaction Problems (CSPs). By systematically relaxing each CSP by removing specific constraints we will produce a series of Cheaper Meta PSs (see Figure 1). The idea is that instead of solving all the CSPs we only need to solve those that have a solution when relaxed. This approach would be beneficial if it saves time in identifying re-usable PS. To determine the best means to relax the CSPs we have proposed building a Case-library, which will suggest dropping specific constraints that correspond to the highest search effort (Nordlander, 2002).
Take a Guided Tour
Try a Demonstration
Version 1.90 (ZIP, 19 KB), including full online documentation, downloadable. For
installation instructions email Tomas
Publication regarding MUSKRAT, A Constraint-Based Approach to the Description & Detection of Fitness-for-Purpose., and Exploration on Relaxation Strategies in Random Binary Constraint Satisfaction Problems together with other relevant documents can be accessed:
http://www.csd.abdn.ac.uk/~tnordlan/Publications.htm (Tomas Eric Nordlander)
http://www.csd.abdn.ac.uk/~sleeman/dhs-publications.html (Derek Sleeman)