Meta-Learning a Dynamical Language Model

Abstract: We consider the task of word-level language modeling and study the possibility of combining hidden-states-based short-term representations with medium-term representations encoded in dynamical weights of a LLM. Our work extends recent experiments on LLMs with dynamically evolving weights by casting the language modeling problem into an online learning-to-learn framework in which a meta-learner is trained by gradient-descent to continuously update a LLM weights.