[论文] Soft Symbol Grounding for Prototypical Concepts

论文概要

研究领域: ML 作者: Marcos Galván-López, Nijesh Upreti, Hiram Calvo, Carlos Aguilar-Ibáñez, Vaishak Belle 发布时间: 2026-09-15 arXiv: 2609.12247

中文摘要

神经符号模型通常仅在最终标签上接受监督，中间概念不可观测。由于许多概念赋值都能与给定标签自洽，训练可能正确预测标签却恢复出错误的概念——这种失效被称为推理捷径。原型网络通过将每个概念锚定到少量标注样本来减少捷径，但现有方法仍通过手工设计、特定任务的可微损失将感知与推理耦合，每个任务都必须重新设计。我们提出 Soft-PNet，移除了这个损失：它将概念落地重新表述为在预计算的可行符号解缓存上的 Metropolis 游走，由每个概念单个标注锚点构建的原型分布引导，并以原型加权缓存与网络概念预测之间的一个 KL 目标进行训练。该目标在所有任务上完全一致，且在解空间无法枚举时仍然适用。在 MNIST-EvenOdd、数独视觉版与 Kand-Logic 的稀缺监督设置下，Soft-PNet 在概念与标签层面均与损失工程化的原型网络持平，并恢复了软落地基线遗漏的概念，且无需损失工程、训练时间更短。

原文摘要

Neuro-symbolic models are usually trained with supervision only on final labels, leaving the intermediate concepts unobserved. Since many concept assignments are consistent with a given label, training can predict labels correctly while recovering the wrong concepts, a failure known as a reasoning shortcut. Prototypical networks reduce shortcuts by anchoring each concept to a few labeled examples, but existing methods still couple perception and reasoning through a hand-crafted, task-specific differentiable loss that must be redesigned for every task. We introduce \textbf{Soft-PNet}, which removes this loss: it reframes concept grounding as a Metropolis walk over a precomputed cache of feasible symbolic solutions, guided by a prototype distribution built from a single labeled anchor per co...

#论文 #arXiv #ML #小凯