How to argue less

(File Last Modified Wed, May 29, 2002.)


Review: How to argue less

Experiments:

  • A random search engine called CHEETAH exercises argument spaces expressed in the JANE rule-based language.
  • TARZAN watches as CHEETAH chases JANE around the argument space.
  • TARZAN builds a log of Jane's behaviour and learns how to nudge JANE into better behaviour.

Conclusion

  • We can argue less by sorting our arguments according to their information gain.
  • We can argue less by applying randomized abduction + induction to our combined viewpoints
  • Funnel theory predicts that most informative arguments are few in number and fast to find.
  • Such most informative arguments can be quickly found in JANE rules via CHEETAH's randomized abduction followed by TARZAN's induction.

Discussion

This technique applies in domains where three factors are true:

  • narrow funnels are frequent
  • random search is an adequate method of searching for goals
  • random search has a preference for narrow funnels over wide funnels

Build 11. Apr 12, 2003


  *  Home

  *  About this site

Literature Review
  *  Data Mining

  *  Machine Learning

  *  Software Engineering

  *  Research Notes



A

argueless.pod
How to argue less


M

mysterious.pod
The Mysterious Case of The Missing Reusable Class Libraries

ml4re.pod
Machine Learning for Requirements Engineering

ml4se.pod
Practical Machine Learning for Software Engineering and Knowledge Engineering


P

pigE.pod
An Expert System for Raising Pigs


R

reasoning.pod
Better reasoning about software engineering activities


W

whatif.pod
Practical Large Scale What-if Queries: Case Studies with Software Risk Assessment