A Deep Learning Approach for Automated Constraint Network Generalization

Read full text Summarise with VIRAT
Authors: Samaksh Chandra, Steven Prestwich, Gregory M. Provan
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Constraint acquisition (CA) methods learn constraint models from examples of solutions and (usually) non-solutions, to automate the difficult task of mathematical modelling. Most CA methods learn constraints for problems of a certain size from examples of that size, but a few can learn from examples of other sizes. Recent work has decomposed this task into (i) learning models of multiple fixed sizes, and (ii) extrapolating these to other sizes. This has the advantage that fixed-size models can be learned by any convenient CA method. In this paper, we a) propose ExCON, a novel, more automated approach (than current methods) for constraint network generalization, b) provide empirical evidence of ExCON's performance on five diverse CA benchmarks, c) compare ExCON with state-of-the-art CSP generalization methods, and d) evaluate ExCON under different noise settings. ExCON is designed to generalize constraint models of unseen sizes by learning from multiple fixed-size instances in a more automated manner. We demonstrate that ExCON performs accurately on this task and shows competitive performance with an approach based on classifiers with feature engineering.

Fields of Study & Subjects

Constraint Satisfaction and Optimization AI-based Problem Solving and Planning Logic Reasoning and Knowledge

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
24 March 2026
Language
EN