Deep Learning Model Selection With Parametric Complexity Control

Olga Grebenkova, Olga Grebenkova, Oleg Bakhteev, Oleg Bakhteev, Vadim Strijov

2023

Abstract

The paper is devoted to deep learning model complexity. It is estimated by Bayesian inference and based on a computational budget. The idea of the proposed method is to represent deep learning model parameters in the form of hypernetwork output. A hypernetwork is a supplementary model which generates parameters of the selected model. This paper considers the minimum description length from a Bayesian point of view. We introduce prior distributions of deep learning model parameters to control the model complexity. The paper analyzes and compares three types of regularization to define the parameter distribution. It infers and generalizes the model evidence as a criterion that depends on the required model complexity. Finally, it analyzes this method in the computational experiments on the Wine, MNIST, and CIFAR-10 datasets.

Paper Citation

in Harvard Style

Grebenkova O., Bakhteev O. and Strijov V. (2023). Deep Learning Model Selection With Parametric Complexity Control. In Proceedings of the 15th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART, ISBN 978-989-758-623-1, pages 65-74. DOI: 10.5220/0011626900003393

in Bibtex Style

@conference{icaart23,
author={Olga Grebenkova and Oleg Bakhteev and Vadim Strijov},
title={Deep Learning Model Selection With Parametric Complexity Control},
booktitle={Proceedings of the 15th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART,},
year={2023},
pages={65-74},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0011626900003393},
isbn={978-989-758-623-1},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 15th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART,
TI - Deep Learning Model Selection With Parametric Complexity Control
SN - 978-989-758-623-1
AU - Grebenkova O.
AU - Bakhteev O.
AU - Strijov V.
PY - 2023
SP - 65
EP - 74
DO - 10.5220/0011626900003393