MetaRuleGPT: Recursive Numerical Reasoning of Language Models Trained with Simple Rules

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chen, Kejie, Wang, Lin, Zhang, Qinghai, Xu, Renjun
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917871937912832
author Chen, Kejie
Wang, Lin
Zhang, Qinghai
Xu, Renjun
author_facet Chen, Kejie
Wang, Lin
Zhang, Qinghai
Xu, Renjun
contents Recent studies have highlighted the limitations of large language models in mathematical reasoning, particularly their inability to capture the underlying logic. Inspired by meta-learning, we propose that models should acquire not only task-specific knowledge but also transferable problem-solving skills. We introduce MetaRuleGPT, a novel Transformer-based architecture that performs precise numerical calculations and complex logical operations by learning and combining different rules. In contrast with traditional training sets, which are heavily composed of massive raw instance data, MetaRuleGPT is pre-trained on much less abstract datasets containing basic, compound, and iterative rules for mathematical reasoning. Extensive experimental results demonstrate MetaRuleGPT can mimic human's rule-following capabilities, break down complexity, and iteratively derive accurate results for complex mathematical problems. These findings prove the potential of rule learning to enhance the numerical reasoning abilities of language models.
format Preprint
id arxiv_https___arxiv_org_abs_2412_13536
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle MetaRuleGPT: Recursive Numerical Reasoning of Language Models Trained with Simple Rules
Chen, Kejie
Wang, Lin
Zhang, Qinghai
Xu, Renjun
Computation and Language
Recent studies have highlighted the limitations of large language models in mathematical reasoning, particularly their inability to capture the underlying logic. Inspired by meta-learning, we propose that models should acquire not only task-specific knowledge but also transferable problem-solving skills. We introduce MetaRuleGPT, a novel Transformer-based architecture that performs precise numerical calculations and complex logical operations by learning and combining different rules. In contrast with traditional training sets, which are heavily composed of massive raw instance data, MetaRuleGPT is pre-trained on much less abstract datasets containing basic, compound, and iterative rules for mathematical reasoning. Extensive experimental results demonstrate MetaRuleGPT can mimic human's rule-following capabilities, break down complexity, and iteratively derive accurate results for complex mathematical problems. These findings prove the potential of rule learning to enhance the numerical reasoning abilities of language models.
title MetaRuleGPT: Recursive Numerical Reasoning of Language Models Trained with Simple Rules
topic Computation and Language
url https://arxiv.org/abs/2412.13536