Compositional Neuro-Symbolic Reasoning

TL;DR

This paper introduces a neuro-symbolic reasoning architecture that combines neural perception with symbolic transformation filtering, significantly improving generalization on the ARC task without task-specific training.

Contribution

It proposes a novel compositional neuro-symbolic framework that enhances language models with object-based reasoning, achieving better generalization on structured reasoning tasks.

Findings

Performance on ARC-AGI-2 improved from 16% to 24.4%.

Combining with ARC Lang Solver increases accuracy to 30.8%.

The approach reduces reliance on brute-force search and sampling.

Abstract

We study structured abstraction-based reasoning for the Abstraction and Reasoning Corpus (ARC) and compare its generalization to test-time approaches. Purely neural architectures lack reliable combinatorial generalization, while strictly symbolic systems struggle with perceptual grounding. We therefore propose a neuro-symbolic architecture that extracts object-level structure from grids, uses neural priors to propose candidate transformations from a fixed domain-specific language (DSL) of atomic patterns, and filters hypotheses using cross-example consistency. Instantiated as a compositional reasoning framework based on unit patterns inspired by human visual abstraction, the system augments large language models (LLMs) with object representations and transformation proposals. On ARC-AGI-2, it improves base LLM performance from 16% to 24.4% on the public evaluation set, and to 30.8% when combined with ARC Lang Solver via a meta-classifier. These results demonstrate that separating perception, neural-guided transformation proposal, and symbolic consistency filtering improves generalization without task-specific finetuning or reinforcement learning, while reducing reliance on brute-force search and sampling-based test-time scaling. We open-source the ARC-AGI-2 Reasoner code (https://github.com/CoreThink-AI/arc-agi-2-reasoner).