Noether Networks: meta-learning useful conserved quantitiesDownload PDF

Published: 09 Nov 2021, Last Modified: 25 Nov 2024NeurIPS 2021 PosterReaders: Everyone
Keywords: meta-learning, sequential prediction, physics, video prediction
TL;DR: We meta-learn inductive biases in the form of useful conserved quantities that improve predictions when optimized inside the prediction function.
Abstract: Progress in machine learning (ML) stems from a combination of data availability, computational resources, and an appropriate encoding of inductive biases. Useful biases often exploit symmetries in the prediction problem, such as convolutional networks relying on translation equivariance. Automatically discovering these useful symmetries holds the potential to greatly improve the performance of ML systems, but still remains a challenge. In this work, we focus on sequential prediction problems and take inspiration from Noether's theorem to reduce the problem of finding inductive biases to meta-learning useful conserved quantities. We propose Noether Networks: a new type of architecture where a meta-learned conservation loss is optimized inside the prediction function. We show, theoretically and experimentally, that Noether Networks improve prediction quality, providing a general framework for discovering inductive biases in sequential problems.
Code Of Conduct: I certify that all co-authors of this work have read and commit to adhering to the NeurIPS Statement on Ethics, Fairness, Inclusivity, and Code of Conduct.
Supplementary Material: pdf
Code: https://lis.csail.mit.edu/noether
Community Implementations: [![CatalyzeX](/images/catalyzex_icon.svg) 1 code implementation](https://www.catalyzex.com/paper/noether-networks-meta-learning-useful/code)
13 Replies

Loading