Abstract
The result of reinforcement learning is often obtained in the form of a q-table mapping actions to future rewards. We propose to use SMT solvers and strategy trees to generate a representation of a learned strategy in a format which is understandable for a human. We present the methodology and demonstrate it on a small game.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
As well as a set of meta-parameters to the learning algorithm, e.g., learning rate.
- 2.
For simplicity, action \(a_2\) is forbidden when \(s = 1\), i.e., it is impossible to pick more sticks than remaining.
- 3.
Where % is the remainder operator.
- 4.
The subtraction of one comes from the action space being defined as \(\{a_1= 0, a_2 = 1\}\) instead of the number of sticks removed (\(\{a_1 = 1, a_2 = 2\}\)).
- 5.
Interval constraints are added on edges, limiting the functions domains for efficiency.
References
Barrett, C., Tinelli, C.: Satisfiability modulo theories. In: Handbook of Model Checking, pp. 305–343. Springer, Cham (2018). https://doi.org/10.1007/978-3-319-10575-8_11
de Moura, L., Bjørner, N.: Z3: an efficient SMT solver. In: Ramakrishnan, C.R., Rehof, J. (eds.) TACAS 2008. LNCS, vol. 4963, pp. 337–340. Springer, Heidelberg (2008). https://doi.org/10.1007/978-3-540-78800-3_24
Sutton, R.S., Barto, A.G.: Reinforcement Learning: An Introduction, 2nd edn. The MIT Press (2018)
Wu, K., et al.: Automatic synthesis of generalized winning strategies of impartial combinatorial games using SMT solvers. In: Bessiere, C. (ed.) Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence, IJCAI-20, pp. 1703–1711. International Joint Conferences on Artificial Intelligence Organization (2020)
Acknowledgements
This work was supported by the Knowledge Foundation in Sweden through the ACICS project (20190038).
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2024 The Author(s), under exclusive license to Springer Nature Switzerland AG
About this paper
Cite this paper
Backeman, P. (2024). Synthesizing Understandable Strategies. In: Kofroň, J., Margaria, T., Seceleanu, C. (eds) Engineering of Computer-Based Systems. ECBS 2023. Lecture Notes in Computer Science, vol 14390. Springer, Cham. https://doi.org/10.1007/978-3-031-49252-5_15
Download citation
DOI: https://doi.org/10.1007/978-3-031-49252-5_15
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-031-49251-8
Online ISBN: 978-3-031-49252-5
eBook Packages: Computer ScienceComputer Science (R0)