When can in-context learning generalize out of task distribution?

TL;DR

This paper empirically investigates the conditions under which in-context learning (ICL) in transformers generalizes out-of-distribution, focusing on task diversity, model depth, and problem dimensionality, revealing a phase transition in generalization capabilities.

Contribution

It introduces a new notion of task diversity and demonstrates its role in enabling transformers to transition from in-distribution to out-of-distribution generalization in ICL.

Findings

Increased task diversity leads to a phase transition in ICL capabilities.

Transformers can generalize out-of-distribution after a critical level of task diversity.

Model depth and problem dimensionality influence the transition point.

Abstract

In-context learning (ICL) is a remarkable capability of pretrained transformers that allows models to generalize to unseen tasks after seeing only a few examples. We investigate empirically the conditions necessary on the pretraining distribution for ICL to emerge and generalize \emph{out-of-distribution}. Previous work has focused on the number of distinct tasks necessary in the pretraining dataset. Here, we use a different notion of task diversity to study the emergence of ICL in transformers trained on linear functions. We find that as task diversity increases, transformers undergo a transition from a specialized solution, which exhibits ICL only within the pretraining task distribution, to a solution which generalizes out of distribution to the entire task space. We also investigate the nature of the solutions learned by the transformer on both sides of the transition, and observe similar transitions in nonlinear regression problems. We construct a phase diagram to characterize how our concept of task diversity interacts with the number of pretraining tasks. In addition, we explore how factors such as the depth of the model and the dimensionality of the regression problem influence the transition.