FSPool: Learning Set Representations with Featurewise Sort Pooling
Machine Learning
2020-05-04 v4 Artificial Intelligence
Machine Learning
Abstract
Traditional set prediction models can struggle with simple datasets due to an issue we call the responsibility problem. We introduce a pooling method for sets of feature vectors based on sorting features across elements of the set. This can be used to construct a permutation-equivariant auto-encoder that avoids this responsibility problem. On a toy dataset of polygons and a set version of MNIST, we show that such an auto-encoder produces considerably better reconstructions and representations. Replacing the pooling function in existing set encoders with FSPool improves accuracy and convergence speed on a variety of datasets.
Cite
@article{arxiv.1906.02795,
title = {FSPool: Learning Set Representations with Featurewise Sort Pooling},
author = {Yan Zhang and Jonathon Hare and Adam Prügel-Bennett},
journal= {arXiv preprint arXiv:1906.02795},
year = {2020}
}
Comments
Published at International Conference on Learning Representations (ICLR) 2020