Abstract
In this paper we model and solve the popular game Wordle using Uppaal Stratego. We model three different game-modes in terms of POMDPs, with more than 12,000 controllable actions. These constitute by far the largest models ever presented to Uppaal Stratego. Our experimental evaluation is encouraging: e.g. in the hard game-mode the partitioning-refinement learning method of Uppaal Stratego reduces the expected number of guesses from a baseline of 7.67 to 4.40 using 1 million training episodes. To better understand the convergence properties of our learning method we also study reduced versions of Wordle.
Originalsprog | Engelsk |
---|---|
Titel | A Journey from Process Algebra via Timed Automata to Model Learning : Essays Dedicated to Frits Vaandrager on the Occasion of His 60th Birthday |
Redaktører | Nils Jansen, Mariëlle Stoelinga, Petra van den Bos |
Antal sider | 23 |
Forlag | Springer |
Publikationsdato | 2022 |
Sider | 283-305 |
ISBN (Trykt) | 978-3-031-15628-1 |
ISBN (Elektronisk) | 978-3-031-15629-8 |
DOI | |
Status | Udgivet - 2022 |
Navn | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
---|---|
Vol/bind | 13560 LNCS |
ISSN | 0302-9743 |
Bibliografisk note
Publisher Copyright:© 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.