用神经符号方法提升模型对抽象推理任务的泛化能力
Compositional Neuro-Symbolic Reasoning
- 从网格中提取物体结构，用神经网络提议变换模式
- 在ARC-AGI-2上将LLM准确率从16%提升至30.8%
- 无需微调或强化学习，适合需要可靠推理的AI系统
我们研究了基于结构抽象的推理方法在抽象与推理基准（ARC）上的泛化表现，并对比其与测试时方法的差异。纯神经架构缺乏可靠的组合泛化能力，而严格符号系统则难以处理感知基础问题。为此，我们提出一种神经符号架构：从网格中提取对象级结构，利用神经先验从固定领域语言（DSL）中提议候选变换，再通过跨例一致性过滤假设。该框架以人类视觉抽象启发的原子模式为基础，增强大语言模型（LLMs）的对象表示与变换提议能力。在ARC-AGI-2上，该系统将基础LLM性能从16%提升至24.4%（公开评估集），结合ARC Lang Solver通过元分类器进一步提升至30.8%。结果表明，分离感知、神经引导变换提议与符号一致性过滤，可在不依赖任务特定微调或强化学习的情况下提升泛化能力，同时减少对暴力搜索和采样式测试时扩展的依赖。代码已开源。
原文摘要 · Abstract (English)
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).
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。