Hamlet Page
Description:
Hamlet is a system that learns control rules (heuristics) for
Prodigy4.0 (now IPSS),
using a lazy and incremental approach. The main goal of the research is to
build a domain-independent learning system that can improve the
efficiency of the problem solving task, as well as the quality
of the solutions provided by a nonlinear planner. The long-discussed fact that
the nonlinear planner searches in the state space instead of in the plan space
does not affect nor search efficiency nor validity of the learning system.
It is composed of two modules: Bounded-Explanation, and Inductive refinement.
The Bounded-Explanation module generates
control rules from the search tree that Prodigy expands on solving a
problem. Those control rules can be either overgeneral or overspecific.
The Inductive module
generalizes those rules, so they can apply in more situations.
The Refinement module specializes the rules,
according to found negative examples, so that they do not apply when they
should not.
The following figure shows the architecture of the system.
The input parameters of the system are shown in the left, where the most
important ones are the set of training problems and the domain. The output is
the set of learned control rules.
You can see some examples of control rules generated by Hamlet for the
classical Blocksworld or the Logistics domains.
How to get and install software:
Example domains:
Related papers (go here for an updated list)
Lazy incremental
learning of control knowledge for efficiently obtaining quality plans,
Daniel Borrajo and Manuela Veloso.
AI Review Journal. Special Issue on Lazy Learning , pages 1-34,
1996.
Integrating planning
and learning: The Prodigy architecture,
Manuela Veloso, Jaime Carbonell, Alicia Pérez, Daniel Borrajo, Eugene
Fink, and Jim Blythe.
Journal of Experimental and Theoretical AI , pages 81-120,
1995.
Authors:
PLTool: a user interative tool for planning and learning download manual (code inside Hamlet already)
and follow instructions