Self-supervised Analogical Learning using Language Models

2025-02-03

AI Summary

Large language models (LLMs) exhibit weak generalization and poor solution consistency on out-of-distribution (OOD) reasoning tasks. To address this, we propose a generalization-enhancement method based on self-supervised analogical learning—the first to formalize human analogical reasoning as a trainable framework. Our approach leverages LMs to automatically extract symbolic solutions, construct cross-problem analogy pairs, and explicitly transfer reasoning paths via contrastive distillation. Crucially, it requires no human annotations, avoids answer memorization, and emphasizes abstraction and reuse of solution structures. Evaluated on benchmarks including StrategyQA, GSM8K, and HotpotQA, our method achieves consistent improvements of 2–20% in OOD generalization accuracy, while significantly enhancing cross-distribution reasoning capability, controllability, and robustness. This work establishes a novel paradigm for symbol-level generalization in foundation models.

Technology Category

• Cognitive Modeling & Cognitive Systems: Analogy
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large language models have been shown to suffer from reasoning inconsistency issues. That is, they fail more in situations unfamiliar to the training data, even though exact or very similar reasoning paths exist in more common cases that they can successfully solve. Such observations motivate us to propose methods that encourage models to understand the high-level and abstract reasoning processes during training instead of only the final answer. This way, models can transfer the exact solution to similar cases, regardless of their relevance to the pre-training data distribution. In this work, we propose SAL, a self-supervised analogical learning framework. SAL mimics the human analogy process and trains models to explicitly transfer high-quality symbolic solutions from cases that they know how to solve to other rare cases in which they tend to fail more. We show that the resulting models after SAL learning outperform base language models on a wide range of reasoning benchmarks, such as StrategyQA, GSM8K, and HotpotQA, by 2% to 20%. At the same time, we show that our model is more generalizable and controllable through analytical studies.

Problem

• Large Language Models
• Generalization Ability
• Problem-Solving Understanding

Innovation

• SAL self-learning method
• Generalization ability enhancement
• Performance improvement on new problems