N-Gram Induction Heads for In-Context RL: Improving Stability and Reducing Data Needs

TL;DR

This paper introduces n-gram induction heads into transformers to enhance in-context reinforcement learning, reducing data requirements and improving training stability, often outperforming existing methods like Algorithm Distillation.

Contribution

The integration of n-gram induction heads into transformers for in-context RL is novel, leading to more data-efficient and stable training compared to prior approaches.

Findings

Reduces data needs for generalization in in-context RL

Eases training sensitivity to hyperparameters

Matches or surpasses Algorithm Distillation performance

Abstract

In-context learning allows models like transformers to adapt to new tasks from a few examples without updating their weights, a desirable trait for reinforcement learning (RL). However, existing in-context RL methods, such as Algorithm Distillation (AD), demand large, carefully curated datasets and can be unstable and costly to train due to the transient nature of in-context learning abilities. In this work, we integrated the n-gram induction heads into transformers for in-context RL. By incorporating these n-gram attention patterns, we considerably reduced the amount of data required for generalization and eased the training process by making models less sensitive to hyperparameters. Our approach matches, and in some cases surpasses, the performance of AD in both grid-world and pixel-based environments, suggesting that n-gram induction heads could improve the efficiency of in-context RL.