Vladislav Meshkov, Nikita Kiselev, Andrey Grabovoy
1 Moscow Institute of Physics and Technology
Installation:
- git clone git@github.com:Drago160/Meshkov-paper.git
- pip install requirements.txt
Experiments:
Choose any config you need and train appropriate group of models, for exmple:
python3 train --config_path=configs/layers_num/fashion_mnist.yml
Visualize it, using
python3 visualize_results.py --config_path=config/layers_num/fashion_mnist.yml
Results will be in figures/fashion_mnist_change_layers.pdf
@INPROCEEDINGS{10899113,
author={Meshkov, Vladislav and Kiselev, Nikita and Grabovoy, Andrey},
booktitle={2024 Ivannikov Ispras Open Conference (ISPRAS)},
title={ConvNets Landscape Convergence: Hessian-Based Analysis of Matricized Networks},
year={2024},
volume={},
number={},
pages={1-10},
keywords={Geometry;Sensitivity;Artificial neural networks;Network architecture;Convergence},
doi={10.1109/ISPRAS64596.2024.10899113}
}