N-Gram Induction Heads for In-Context RL: Improving Stability and Reducing Data Needs
This work addresses efficiency issues for researchers and practitioners using in-context RL, though it is incremental as it builds on existing methods like Algorithm Distillation.
The paper tackled the problem of instability and high data requirements in in-context reinforcement learning (RL) by integrating n-gram induction heads into transformers, resulting in reduced data needs and improved training stability while matching or surpassing the performance of Algorithm Distillation in grid-world and pixel-based environments.
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.