Complexity Control Facilitates Reasoning-Based Compositional Generalization in Transformers

2025-01-15

AI Summary

It remains unclear whether Transformers achieve compositional generalization through rule-based reasoning or merely via memorization of specific instances. Method: We introduce a complexity-control framework grounded in low-complexity bias as the key mechanism enabling reasoning-driven generalization. Leveraging information-path masking, multi-dimensional complexity metrics, and cross-modal interpretability analysis (spanning image generation and NLP), we establish, for the first time, a theoretical link between complexity modulation and neuronal condensation. Our strategy explicitly steers models toward learning transferable primitive rules rather than superficial patterns. Contribution/Results: Evaluated across multiple real-world datasets, our approach consistently enhances out-of-distribution compositional generalization performance. It provides a novel paradigm for understanding the generalization mechanisms of large language models, bridging formal complexity theory with empirical neural behavior.

Technology Category

• Natural Language Processing: (Large) Language Models
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Transformers have demonstrated impressive capabilities across various tasks, yet their performance on compositional problems remains a subject of debate. In this study, we investigate the internal mechanisms underlying Transformers' behavior in compositional tasks. We find that complexity control strategies significantly influence whether the model learns primitive-level rules that generalize out-of-distribution (reasoning-based solutions) or relies solely on memorized mappings (memory-based solutions). By applying masking strategies to the model's information circuits and employing multiple complexity metrics, we reveal distinct internal working mechanisms associated with different solution types. Further analysis reveals that reasoning-based solutions exhibit a lower complexity bias, which aligns with the well-studied neuron condensation phenomenon. This lower complexity bias is hypothesized to be the key factor enabling these solutions to learn reasoning rules. We validate these conclusions across multiple real-world datasets, including image generation and natural language processing tasks, confirming the broad applicability of our findings.

Problem

• Transformer
• Complex Problem Solving
• Learning Mechanism

Innovation

• Complexity Control
• Neuronal Pruning Phenomenon
• Transformer Optimization