arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2604.02434cs.AI
收藏

组合式神经符号推理

Compositional Neuro-Symbolic Reasoning

CoreThink AI；Stanford University（斯坦福大学）

Anugyan Das, Omkar Ghugarkar, Vishvesh Bhat, Asad Aali

发表信息：
TMLR, 2026
阅读 PDF
arXiv 页面 ↗HTML 原文 ↗

AI总结：

本文提出一种结合神经和符号方法的推理框架，通过提取网格对象结构、利用神经先验提出变换候选方案并过滤假设，提升ARC-AGI-2任务表现，无需任务特定微调或强化学习。

摘要：

我们研究了基于结构抽象的推理方法用于Abstraction and Reasoning Corpus (ARC)，并将其泛化能力与测试时方法进行比较。纯神经架构缺乏可靠的组合泛化能力，而严格符号系统在感知接地方面遇到困难。因此，我们提出了一种神经符号架构，从网格中提取对象级结构，利用神经先验提出固定领域特定语言（DSL）中原子模式的候选变换，并通过跨示例一致性过滤假设。该系统作为基于单元模式的组合推理框架，灵感来自人类视觉抽象，增强了大型语言模型（LLMs）的对象表示和变换提案。在ARC-AGI-2上，它将基础LLM性能从16%提升到24.4%，并在结合ARC Lang Solver时提升至30.8%。这些结果表明，分离感知、神经引导的变换提案和符号一致性过滤能提升泛化能力，无需任务特定微调或强化学习，同时减少对暴力搜索和基于采样测试时扩展的依赖。我们开源了ARC-AGI-2 Reasoner代码（https://github.com/CoreThink-AI/arc-agi-2-reasoner）。

英文摘要：

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 checks. 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 over the base solver LLM (Grok-4, 16.0%) to 24.4% on the public evaluation set, and to 30.8% when combined with ARC Lang Solver via a meta-classifier. These results support the claim that in the evaluated settings, allocating test-time compute to object abstraction, neural-guided transformation proposal, and consistency filtering can improve accuracy over unstructured sampling-based scaling without task-specific finetuning of the system. We open-source the code: (https://github.com/CoreThink-AI/arc-agi-2-reasoner).

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。

↑