Meta-Learning Symmetries by Reparameterization

07/06/2020
by   Allan Zhou, et al.
5

Many successful deep learning architectures are equivariant to certain transformations in order to conserve parameters and improve generalization: most famously, convolution layers are equivariant to shifts of the input. This approach only works when practitioners know a-priori symmetries of the task and can manually construct an architecture with the corresponding equivariances. Our goal is a general approach for learning equivariances from data, without needing prior knowledge of a task's symmetries or custom task-specific architectures. We present a method for learning and encoding equivariances into networks by learning corresponding parameter sharing patterns from data. Our method can provably encode equivariance-inducing parameter sharing for any finite group of symmetry transformations, and we find experimentally that it can automatically learn a variety of equivariances from symmetries in data. We provide our experiment code and pre-trained models at https://github.com/AllanYangZhou/metalearning-symmetries.

READ FULL TEXT

page 6

page 14

research
03/20/2023

Architecture, Dataset and Model-Scale Agnostic Data-free Meta-Learning

The goal of data-free meta-learning is to learn useful prior knowledge f...
research
11/18/2022

Task Residual for Tuning Vision-Language Models

Large-scale vision-language models (VLMs) pre-trained on billion-level d...
research
07/04/2023

Learning Lie Group Symmetry Transformations with Neural Networks

The problem of detecting and quantifying the presence of symmetries in d...
research
02/24/2022

KESA: A Knowledge Enhanced Approach For Sentiment Analysis

Though some recent works focus on injecting sentiment knowledge into pre...
research
04/05/2019

Branched Multi-Task Networks: Deciding What Layers To Share

In the context of deep learning, neural networks with multiple branches ...
research
10/28/2017

Multi-Task Learning by Deep Collaboration and Application in Facial Landmark Detection

Convolutional neural networks (CNN) have become the most successful and ...
research
04/24/2022

Learning Symmetric Embeddings for Equivariant World Models

Incorporating symmetries can lead to highly data-efficient and generaliz...

Please sign up or login with your details

Forgot password? Click here to reset