The Buffer Mechanism for Multi-Step Information Reasoning in Language Models

2024-05-24

AI Summary

Large language models exhibit limited performance on multi-step symbolic reasoning tasks, such as mathematical problem solving. To address this, we introduce a novel “buffer-based” paradigm grounded in symbolic multi-step reasoning data construction and Transformer attention mechanism analysis. Our approach decouples vertical (inter-layer) and horizontal (inter-step) reasoning pathways, enabling dynamic information storage and selective retrieval. We further propose a query-key-driven buffer modeling framework and a stochastic matrix optimization algorithm to enhance information scheduling efficiency. Evaluated on the PrOntoQA benchmark, our method reduces GPT-2’s generalization training time by 75%. Beyond empirical gains, this work identifies critical information propagation bottlenecks in chain-of-thought reasoning and establishes an interpretable, scalable framework for enhancing multi-step symbolic reasoning in neural language models.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Large language models have consistently struggled with complex reasoning tasks, such as mathematical problem-solving. Investigating the internal reasoning mechanisms of these models can help us design better model architectures and training strategies, ultimately enhancing their reasoning capability. In this study, we constructed a symbolic dataset to investigate the mechanisms by which Transformer models employ vertical thinking strategy based on their inherent structure and horizontal thinking strategy based on Chain of Thought to achieve multi-step reasoning. We introduced the concept of buffer mechanism: the model stores various information in distinct buffers and selectively extracts them through the query-key matrix. We proposed a random matrix-based algorithm to enhance the model's reasoning ability, resulting in a 75% reduction in the training time required for the GPT-2 model to achieve generalization capability on the PrOntoQA dataset. These findings provide new insights into understanding the mechanisms of large language models.

Problem

• Investigating internal reasoning mechanisms of large language models
• Enhancing model performance on multi-step symbolic reasoning tasks
• Developing buffer mechanism for information storage and extraction

Innovation

• Introduced buffer mechanism for information storage
• Developed random matrix-based reasoning algorithm
• Added minimal trainable parameters for performance boost