Learn Smarter, Not Harder: Improving Uppaal Stratego through Preprocessing

Terndrup, Kasper Kohsel ; Sillesen, Simon Vandel

4. term

Software, Master

2018

2018-05-31

97 pages

The UPPAAL STRATEGO tool can synthesize near-optimal strategies for Priced Timed Markov Decision Processes. However, model elements that are irrelevant or redundant for the optimal strategy, can mislead the synthesis by needlessly increasing the state space. In this thesis, we propose a preprocessing addition to the UPPAAL STRATEGO algorithm, that can provide relief for redundancy and irrelevance in the synthesis. The addition enables the application of Principal Component Analysis or Fast Correlation Based Filter with the intention of reducing or removing irrelevant and redundant elements from models. We conduct a series of experiments, and show that preprocessing can improve strategy synthesis, in terms of better strategy performance and reduced size of the produced strategies. The results provide a basis for the inclusion of preprocessing capabilities, in the future development of UPPAAL STRATEGO.

uppaal ; stratego ; strategy ; preprocessing ; reduction ; ptmdp ; timed automata ; markov decision process ; fast correlation based filter ; principal component analysis ; synthesis ; machine learning ; feature irrelevance ; feature redundancy ; feature selection ; feature extraction

Download
View record in AAU Student Projects