Fundamental Reasoning Paradigms Induce Out-of-Domain Generalization in Language Models

TL;DR

This paper explores how fundamental reasoning paradigms like deduction, induction, and abduction influence large language models' ability to generalize out-of-domain, using a new dataset and various training methods to enhance reasoning skills.

Contribution

It introduces a dataset of reasoning trajectories for core paradigms and systematically evaluates methods to induce these reasoning skills into LLMs for better out-of-domain generalization.

Findings

Inducing reasoning paradigms improves out-of-domain performance by up to 14.60 points.

Different training approaches, including fine-tuning and model transformation, effectively enhance reasoning capabilities.

The study demonstrates the importance of fundamental reasoning paradigms for robust LLM generalization.

Abstract

Deduction, induction, and abduction are fundamental reasoning paradigms, core for human logical thinking. Although improving Large Language Model (LLM) reasoning has attracted significant research efforts, the extent to which the fundamental paradigms induce generalization has yet to be systematically explored. In this study, we shed light on how the interplay between these core paradigms influences LLMs' reasoning behavior. To this end, we first collect a new dataset of reasoning trajectories from symbolic tasks, each targeting one of the three fundamental paradigms, to abstract from concrete world knowledge. Then, we investigate effective ways for inducing these skills into LLMs. We experiment with a battery of methods including simple fine-tuning, and more complex approaches to increase model depth, or transform a dense model to a mixture-of-experts. We comprehensively evaluate induced models on realistic out-of-domain tasks, that are entirely formulated in natural language and contain real-world knowledge. Our results reveal that our approach yields strong generalizability with substantial performance gains (up to $14.60$) across realistic tasks.