Neuro-Symbolic Constraint Acquisition: Learning Symbolic Constraint Representations from Examples with Neural Oracle Transformers

Forum topic · 小凯 · 2026-09-15

Summary

This paper introduces a neuro-symbolic framework for automatic constraint acquisition (CA), aiming to learn user-defined concepts as constraint networks without costly human interaction. Traditional CA methods depend heavily on queries to a human oracle, which is expensive in time and number of queries. The proposed approach trains neural Oracle Transformer models on previously available examples to emulate user responses and generalize conceptual knowledge. The learned oracle then interacts with FastCA, a dedicated CA engine that systematically refines the oracle's answers into a sound, consistent, and interpretable constraint network. By replacing the human oracle with a learned one, the framework recovers structured symbolic models from data without requiring prior domain knowledge. The authors, Nassim Belmecheri, Arnaud Gotlieb, Nadjib Lazaar, and Helge Spieker, argue that this neuro-symbolic collaboration effectively aligns data-driven pattern recognition with symbolic reasoning, offering a robust path toward automatic model construction for combinatorial optimization. The work is available on arXiv as 2609.12267.

论文概要 (Paper Overview)

Research area: Machine Learning Authors: Nassim Belmecheri, Arnaud Gotlieb, Nadjib Lazaar, Helge Spieker Published: 2026-09-15 arXiv: 2609.12267

English Translation (Abstract)

Learning user-defined concepts as constraint networks has been extensively studied in the constraint acquisition (CA) literature. However, existing approaches typically rely on intensive interactions with a human oracle, making the learning process costly in terms of time and number of queries.

In this paper, the authors propose a neuro-symbolic framework for automatic CA that significantly reduces user involvement by introducing neural Oracle Transformer models which learn to emulate user responses and to generalize conceptual knowledge. Trained on previously available examples, the learned oracle interacts with a dedicated CA engine, FastCA, which systematically refines the oracle's responses into a sound, consistent, and interpretable constraint network. This neuro-symbolic interaction enables the recovery of structured symbolic models from data without prior domain knowledge.

The results show that this neuro-symbolic collaboration effectively aligns data-driven pattern recognition with symbolic reasoning, providing a robust path toward automatic model construction in the field of combinatorial optimization.

Key Points

Problem: Constraint acquisition traditionally requires intensive interaction with a human oracle, which is expensive in both time and query count.
Approach: Neural Oracle Transformer models are trained on existing examples to emulate user responses and generalize conceptual knowledge.
System: The learned oracle works with FastCA, a dedicated CA engine that refines responses into sound, consistent, and interpretable constraint networks.
Outcome: The framework recovers structured symbolic models from data without prior domain knowledge, bridging data-driven pattern recognition and symbolic reasoning for automatic model construction in combinatorial optimization.

---

*Auto-collected on 2026-09-15*