ICML 2020poster26 citations
Multidimensional Shape Constraints
Maya Gupta, Erez Louidor, Oleksandr Mangylov, Nobu Morioka, Taman Narayan, Sen Zhao
Abstract
We propose new multi-input shape constraints across four intuitive categories: complements, diminishers, dominance, and unimodality constraints. We show these shape constraints can be checked and even enforced when training machine-learned models for linear models, generalized additive models, and the nonlinear function class of multi-layer lattice models. Real-world experiments illustrate how the different shape constraints can be used to increase explainability and improve regularization, especially for non-IID train-test distribution shift.
BibTeX
@InProceedings{pmlr-v119-gupta20b,
title = {Multidimensional Shape Constraints},
author = {Gupta, Maya and Louidor, Erez and Mangylov, Oleksandr and Morioka, Nobu and Narayan, Taman and Zhao, Sen},
booktitle = {Proceedings of the 37th International Conference on Machine Learning},
pages = {3918--3928},
year = {2020},
editor = {III, Hal Daumé and Singh, Aarti},
volume = {119},
series = {Proceedings of Machine Learning Research},
month = {13--18 Jul},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v119/gupta20b/gupta20b.pdf},
url = {https://proceedings.mlr.press/v119/gupta20b.html},
abstract = {We propose new multi-input shape constraints across four intuitive categories: complements, diminishers, dominance, and unimodality constraints. We show these shape constraints can be checked and even enforced when training machine-learned models for linear models, generalized additive models, and the nonlinear function class of multi-layer lattice models. Real-world experiments illustrate how the different shape constraints can be used to increase explainability and improve regularization, especially for non-IID train-test distribution shift.}
}