arXiv cs.AI· Saurabh Mathur, Sahil Sidheekh, Bhavan Vasu, Farbod Tavakkoli, Prasad Tadepalli, Kristian Kersting, Sriraam Natarajan·· 9 天前AI 评分22
grasp：用函数梯度引导语言模型学习概率逻辑程序
Learning Probabilistic Logic Programs with Functional Gradient Guided Language Models
AI 导读
研究者提出神经符号框架 grasp（Gradient-boosted Synthesis of Probabilistic logic programs），将关系结构学习建模为函数梯度提升，把弱学习器设为一阶规则，并将难以求解的内部搜索交给 LLM 作为提议预言机。
正文 · 原文
Abstract:Declarative logic programs offer a powerful and interpretable abstraction for encoding relational structure and neurosymbolic reasoning, by expressing dependencies as weighted compositional rules. However, inducing them from data remains fundamentally hard, bottlenecked by the combinatorial explosion of symbolic search spaces. LLMs have recently emerged as powerful hypothesis generators, but when used in isolation, they lack the capacity to do systematic inductive reasoning needed to reliably synthesize valid programs that fit complex relational distributions. We introduce grasp (Gradient-boosted Synthesis of Probabilistic logic programs), a neurosymbolic framework that casts relational structure learning as functional gradient boosting in which the weak learner is a first-order rule and the intractable inner search is delegated to an LLM proposal oracle. We evaluate grasp on four relational benchmarks spanning molecular toxicity prediction (Tox21), mutagenesis, and citation matching (Cora), and show that it improves over purely symbolic, neural, and LLM-based baselines, while producing interpretable weighted rule ensembles. By replacing combinatorial search with gradient-guided LLM hypothesis generation, grasp retains boosting guarantees without sacrificing the transparency of symbolic outputs.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.12303 [cs.AI] |
| (or arXiv:2610.12303v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.12303 arXiv-issued DOI via DataCite (pending registration) |
|
| Journal reference: | 6th International Joint Conference on Learning & Reasoning (IJCLR 2026) 6th International Joint Conference on Learning & Reasoning 6th International Joint Conference on Learning & Reasoning (IJCLR) |
Submission history
From: Sahil Sidheekh [view email]
[v1]
Thu, 8 Oct 2026 16:53:32 UTC (311 KB)
来源：arXiv cs.AI · arxiv.org