Machine Learning for Adaptive Planning

Abstract

This chapter is concerned with the enhancement of planning systems using techniques from Machine Learning in order to automatically configure their planning parameters according to the morphology of the problem in hand. It presents two different adaptive systems that set the planning parameters of a highly adjustable planner based on measurable characteristics of the problem instance. The planners have acquired their knowledge from a large data set produced by results from experiments on many problems from various domains. The first planner is a rulebased system that employs propositional rule learning to induce knowledge that suggests effective configuration of planning parameters based on the problem's characteristics. The second planner employs instance-based learning in order to find problems with similar structure and adopt the planner configuration that has proved in the past to be effective on these problems. The validity of the two adaptive systems is assessed through experimental results that demonstrate the boost in performance in problems of both known and unknown domains. Comparative experimental results for the two planning systems are presented along with a discussion of their advantages and disadvantages.

Machine Learning in Hybrid Hierarchical and Partial-Order Planners for Manufacturing Domains

“…A recent application of machine learning and rule-based techniques on planning has been done to build an adaptive planning system, called HAP that can automatically fine-tune its parameters based on the values of specific measurable characteristics of each problem [64]. Adaptation is guided by a rule-based system, whose knowledge has been acquired through machine learning techniques.…”

Machine Learning for Adaptive Planning

“…The authors have worked during the past few years in exploiting Machine Learning techniques for Adaptive Planning and have developed two systems that are described in detail later in this chapter. The first system, called HAP RC (Vrakas et al, 2003a ;2003b), is capable of automatically fine-tuning its planning parameters based on the morphology of the problem in hand. The tuning of HAP RC is performed by a rule system, the knowledge of which has been induced through the application of a classification algorithm over a large dataset containing performance data of past executions of HAP (Highly Adjustable Planner).…”

“…Mining association rules from the resulting dataset was a first idea, which however was turned down due to the fact that it would produce too many rules making it extremely difficult to produce all the relevant ones. In our previous work (Vrakas et al, 2003a), we have used the approach of classification based on association rules (Liu, Hsu & Ma, 1998), which induces association rules that only have a specific target attribute on the right hand side. However, such an approach was proved inappropriate for our current much more extended dataset.…”

A Visualization Environment for Planning

“…The automatic configuration is based on HAP-RC 26 , which uses a rule system in order to automatically select the best settings for each planning parameter, based on the morphology of the problem in hand. HAP-RC, whose architecture is outlined in Figure 7 is actually HAP with two additional modules (Problem Analyzer and Rule System) which are utilized off-line, just after reading the representation of the problem in order to fine tune the planning parameters of HAP.…”