B. Satzger and O. Kramer (USA)
AI, Planning, Machine Learning, Linear Regression, Neural Networks
Over the last decades, the complexity of computer systems has increased rapidly. The control of highly complex systems represents a tough challenge. Future systems should be able to dynamically and autonomously adapt to their environment according to high-level goals. Automated planning is known to be a vehicle for developing self-organizing, goal-driven computer systems. Planning-based control tends to result in repeatedly occurring planning tasks within the same domain. In such cases it would be desirable to extract information from successfully solved problems, yielding improved planning performance over time. Planning entities would become more experienced and could adapt to the planning and controlling tasks they are faced with. Many successful planning algorithms perform a heuristic search. This paper investigates heuristic function learning for automated planning using linear regression and neural networks. Experiments show that the proposed techniques can be used both for improving the quality of traditional planning heuristics and for building computationally lightweight adaptive heuristics.
Important Links:
Go Back
