Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This study investigates how Large Language Models (LLMs) utilize attention heads to perform logical reasoning by aligning reasoning steps with token logits within a symbolic-aided Chain-of-Thought (CoT) prompting framework. The findings reveal that only a small fraction of attention heads (approximately 3%) are specialized for retrieving factual and rule-based information for sub-reasoning tasks, while higher layers are crucial for integrating information across multiple reasoning steps. This nuanced understanding of attention head functionality enhances our comprehension of LLMs' reasoning processes and their ability to execute complex tasks through structured algorithms.
LLMs leverage just 3% of their attention heads for specific reasoning tasks, revealing a surprising specialization in their logical reasoning capabilities.
Recent studies have shown that Large Language Models (LLMs) can achieve strong reasoning performance by incorporating functional symbolic representations that abstractly describe graph traversal algorithms and step-by-step reasoning in few-shot learning settings. However, it remains unclear how LLMs genuinely understand the abstract meaning of each reasoning step and the overall algorithm from only a limited number of demonstrations. This work aims to localize the attention heads responsible for individual reasoning steps and characterize the types of information transferred among them. We first align constituent reasoning steps with their corresponding token logits under a symbolic-aided Chain-of-Thought (CoT) prompting framework. Our analysis shows that token positions that steer the reasoning process are associated with low confidence scores caused by constraints on satisfying reasoning behavior patterns in demonstrations. We then adopt causal mediation analysis techniques to identify the attention heads responsible for these patterns. In addition, our findings indicate that LLMs retrieve factual and rule-based information for individual sub-reasoning tasks through specialized attention heads (approximately 3% total heads), whereas higher layers predominantly facilitate information integration and the emergence of global reasoning strategies (e.g., graph traversal algorithms) that coordinate multiple intermediate reasoning steps to solve the overall task.