Unlocking In-Context Learning for Natural Datasets Beyond Language Modelling
Abstract
Large Language Models (LLMs) exhibit In-Context Learning (ICL), which enables the model to perform new tasks conditioning only on the examples provided in the context without updating the model's weights. While ICL offers fast adaptation across natural language tasks and domains, its emergence is less straightforward for modalities beyond text. In this work, we systematically uncover properties present in LLMs that support the emergence of ICL for autoregressive models and various modalities by promoting the learning of the needed mechanisms for ICL. We identify exact token repetitions in the training data sequences as an important factor for ICL. Such repetitions further improve stability and reduce transiency in ICL performance. Moreover, we emphasise the significance of training task difficulty for the emergence of ICL. Finally, by applying our novel insights on ICL emergence, we unlock ICL capabilities for various visual datasets and a more challenging EEG classification task.
Cite
@article{arxiv.2501.06256,
title = {Unlocking In-Context Learning for Natural Datasets Beyond Language Modelling},
author = {Jelena Bratulić and Sudhanshu Mittal and David T. Hoffmann and Samuel Böhm and Robin Tibor Schirrmeister and Tonio Ball and Christian Rupprecht and Thomas Brox},
journal= {arXiv preprint arXiv:2501.06256},
year = {2025}
}
Comments
Best Paper Honorable Mention at GCPR 2025 (German Conference on Pattern Recognition). This is the updated version submitted to the conference, not the official conference proceedings