Dissecting Chain-of-Thought: Compositionality through In-Context Filtering and Learning

TL;DR

This paper investigates how chain-of-thought (CoT) enhances transformer models' ability to learn complex compositional functions by breaking down reasoning into steps, reducing sample complexity, and accelerating pretraining.

Contribution

It provides both experimental and theoretical insights into CoT's mechanics, showing how it improves in-context learning and facilitates learning complex functions.

Findings

CoT reduces sample complexity in in-context learning.

Transformers can learn compositional functions by adding data-filtering layers.

CoT accelerates pretraining by enabling shortcuts for complex functions.

Abstract

Chain-of-thought (CoT) is a method that enables language models to handle complex reasoning tasks by decomposing them into simpler steps. Despite its success, the underlying mechanics of CoT are not yet fully understood. In an attempt to shed light on this, our study investigates the impact of CoT on the ability of transformers to in-context learn a simple to study, yet general family of compositional functions: multi-layer perceptrons (MLPs). In this setting, we find that the success of CoT can be attributed to breaking down in-context learning of a compositional function into two distinct phases: focusing on and filtering data related to each step of the composition and in-context learning the single-step composition function. Through both experimental and theoretical evidence, we demonstrate how CoT significantly reduces the sample complexity of in-context learning (ICL) and facilitates the learning of complex functions that non-CoT methods struggle with. Furthermore, we illustrate how transformers can transition from vanilla in-context learning to mastering a compositional function with CoT by simply incorporating additional layers that perform the necessary data-filtering for CoT via the attention mechanism. In addition to these test-time benefits, we show CoT helps accelerate pretraining by learning shortcuts to represent complex functions and filtering plays an important role in this process. These findings collectively provide insights into the mechanics of CoT, inviting further investigation of its role in complex reasoning tasks.