We discuss the discontinuities that arise when mapping unordered objects to neural network outputs of fixed permutation, referred to as the responsibility problem. Prior work has proved the existence of the issue by identifying a single discontinuity. Here, we show that discontinuities under such models are uncountably infinite, motivating further research into neural networks for unordered data.
Cite
@article{arxiv.2304.09499,
title = {The Responsibility Problem in Neural Networks with Unordered Targets},
author = {Ben Hayes and Charalampos Saitis and György Fazekas},
journal= {arXiv preprint arXiv:2304.09499},
year = {2023}
}
Comments
Accepted for TinyPaper archival at ICLR 2023: https://openreview.net/forum?id=jd7Hy1jRiv4