MC-DARTS (Model Size Constrained Differentiable Architecture Search) adding constraints to DARTS to search for a network architecture considering the accuracy and the model size.
MC-DARTS has been accepted at NeurIPS 2022 Has it Trained Yet? A Workshop for Algorithmic Efficiency in Practical Neural Network Training! MC-DARTS : Model Size Constrained Differentiable Architecture Search
This code was created based on the DARTS. New branches for loss and update formulas have been added (Detailed modifications ).
Python (3.7)
Numpy (1.21.5)
Pytorch (1.5.0)
tqdm (4.64.1)
torchsummary (1.5.1)
python-graphviz (0.20)
python train_search.py --limit_param 2500000
The paper prepares it in advance, but this code downloads the dataset so that it can be run immediately.
python train.py --arch "mc_darts2900000"
See genotypes.py
If you use our code, please cite our paper.
@inproceedings{
hemmi2022mcdarts,
title={{MC}-{DARTS} : Model Size Constrained Differentiable Architecture Search},
author={Kazuki Hemmi and Yuki Tanigaki and Masaki Onishi},
booktitle={Has it Trained Yet? NeurIPS 2022 Workshop},
year={2022},
url={https://openreview.net/forum?id=jKJ6OcvqdQ}
}