Drop in replacements for pytorch nn.Linear for stable learning and inductive priors in physics informed machine learning applications.
$ conda create -n slim python=3.7
$ conda install pytorch torchvision cudatoolkit=10.2 -c pytorch
$ pip install python-mnist@article{SLiM2022,
title={{SLiM: Structured Linear Maps}},
author={Tuor, Aaron and Drgona, Jan and Skomski, Mia},
Url= {https://github.com/pnnl/neuromancer},
year={2022}
}@inproceedings{NEURIPS2021_c9dd73f5,
author = {Drgona, Jan and Mukherjee, Sayak and Zhang, Jiaxin and Liu, Frank and Halappanavar, Mahantesh},
booktitle = {Advances in Neural Information Processing Systems},
editor = {M. Ranzato and A. Beygelzimer and Y. Dauphin and P.S. Liang and J. Wortman Vaughan},
pages = {24033--24047},
publisher = {Curran Associates, Inc.},
title = {On the Stochastic Stability of Deep Markov Models},
url = {https://proceedings.neurips.cc/paper/2021/file/c9dd73f5cb96486f5e1e0680e841a550-Paper.pdf},
volume = {34},
year = {2021}
}