On Deep Set Learning and the Choice of Aggregations
Machine Learning
2020-04-09 v2 Machine Learning
Abstract
Recently, it has been shown that many functions on sets can be represented by sum decompositions. These decompositons easily lend themselves to neural approximations, extending the applicability of neural nets to set-valued inputs---Deep Set learning. This work investigates a core component of Deep Set architecture: aggregation functions. We suggest and examine alternatives to commonly used aggregation functions, including learnable recurrent aggregation functions. Empirically, we show that the Deep Set networks are highly sensitive to the choice of aggregation functions: beyond improved performance, we find that learnable aggregations lower hyper-parameter sensitivity and generalize better to out-of-distribution input size.
Keywords
Cite
@article{arxiv.1903.07348,
title = {On Deep Set Learning and the Choice of Aggregations},
author = {Maximilian Soelch and Adnan Akhundov and Patrick van der Smagt and Justin Bayer},
journal= {arXiv preprint arXiv:1903.07348},
year = {2020}
}