Theoretical and Experimental Results for Planning with Learned Binarized Neural Network Transition Models
(2020) 26th International Conference on Principles and Practice of Constraint Programming, CP 2020 In Lecture Notes in Computer Science 12333. p.917-934- Abstract
We study planning problems where the transition function is described by a learned binarized neural network (BNN). Theoretically, we show that feasible planning with a learned BNN model is NP-complete, and present two new constraint programming models of this task as a mathematical optimization problem. Experimentally, we run solvers for constraint programming, weighted partial maximum satisfiability, 0–1 integer programming, and pseudo-Boolean optimization, and observe that the pseudo-Boolean solver outperforms previous approaches by one to two orders of magnitude. We also investigate symmetry handling for planning problems with learned BNNs over long horizons. While the results here are less clear-cut, we see that exploiting... (More)
We study planning problems where the transition function is described by a learned binarized neural network (BNN). Theoretically, we show that feasible planning with a learned BNN model is NP-complete, and present two new constraint programming models of this task as a mathematical optimization problem. Experimentally, we run solvers for constraint programming, weighted partial maximum satisfiability, 0–1 integer programming, and pseudo-Boolean optimization, and observe that the pseudo-Boolean solver outperforms previous approaches by one to two orders of magnitude. We also investigate symmetry handling for planning problems with learned BNNs over long horizons. While the results here are less clear-cut, we see that exploiting symmetries can sometimes reduce the running time of the pseudo-Boolean solver by up to three orders of magnitude.
(Less)
- author
- Say, Buser ; Devriendt, Jo LU ; Nordström, Jakob LU and Stuckey, Peter J.
- organization
- publishing date
- 2020
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- keywords
- Automated planning, Binarized neural networks, Cutting planes reasoning, Mathematical optimization, Pseudo-Boolean optimization, Symmetry
- host publication
- Principles and Practice of Constraint Programming - 26th International Conference, CP 2020, Proceedings
- series title
- Lecture Notes in Computer Science
- editor
- Simonis, Helmut
- volume
- 12333
- pages
- 18 pages
- publisher
- Springer
- conference name
- 26th International Conference on Principles and Practice of Constraint Programming, CP 2020
- conference location
- Louvain-la-Neuve, Belgium
- conference dates
- 2020-09-07 - 2020-09-11
- external identifiers
-
- scopus:85091310124
- ISSN
- 1611-3349
- 0302-9743
- ISBN
- 9783030584740
- DOI
- 10.1007/978-3-030-58475-7_53
- language
- English
- LU publication?
- yes
- id
- dfe6a76b-540c-4526-b038-f42c7e91cd08
- date added to LUP
- 2020-10-28 14:28:53
- date last changed
- 2024-06-27 00:01:36
@inproceedings{dfe6a76b-540c-4526-b038-f42c7e91cd08,
abstract = {{<p>We study planning problems where the transition function is described by a learned binarized neural network (BNN). Theoretically, we show that feasible planning with a learned BNN model is NP-complete, and present two new constraint programming models of this task as a mathematical optimization problem. Experimentally, we run solvers for constraint programming, weighted partial maximum satisfiability, 0–1 integer programming, and pseudo-Boolean optimization, and observe that the pseudo-Boolean solver outperforms previous approaches by one to two orders of magnitude. We also investigate symmetry handling for planning problems with learned BNNs over long horizons. While the results here are less clear-cut, we see that exploiting symmetries can sometimes reduce the running time of the pseudo-Boolean solver by up to three orders of magnitude.</p>}},
author = {{Say, Buser and Devriendt, Jo and Nordström, Jakob and Stuckey, Peter J.}},
booktitle = {{Principles and Practice of Constraint Programming - 26th International Conference, CP 2020, Proceedings}},
editor = {{Simonis, Helmut}},
isbn = {{9783030584740}},
issn = {{1611-3349}},
keywords = {{Automated planning; Binarized neural networks; Cutting planes reasoning; Mathematical optimization; Pseudo-Boolean optimization; Symmetry}},
language = {{eng}},
pages = {{917--934}},
publisher = {{Springer}},
series = {{Lecture Notes in Computer Science}},
title = {{Theoretical and Experimental Results for Planning with Learned Binarized Neural Network Transition Models}},
url = {{http://dx.doi.org/10.1007/978-3-030-58475-7_53}},
doi = {{10.1007/978-3-030-58475-7_53}},
volume = {{12333}},
year = {{2020}},
}