Google Research

Learning Differentiable Grammars for Continuous Data

CoRR (2019)

Abstract

This paper proposes a novel algorithm which learns the rules of a regular grammar from continuous data, such as video or other streaming data. It is fully differentiable for end-to-end learning and provides easily interpretable results which are important to understand the learned structures. It outperforms the state-of-the-art on several challenging datasets and is also shown to be more accurate for forecasting the future activities in videos. We will open source the code.

Learn more about how we do research

We maintain a portfolio of research projects, providing individuals and teams the freedom to emphasize specific types of work