Aug 12 – 16, 2024
Von-Melle-Park 8
Europe/Berlin timezone

Dissipativity Properties in Neural Network Training

Aug 13, 2024, 10:00 AM
30m
Seminarraum 205 (Von-Melle-Park 8)

Seminarraum 205

Von-Melle-Park 8

Minisymposium Contribution MS 01: Optimal Control and Machine Learning MS 01: Optimal Control and Machine Learning

Speaker

Jens Püttschneider (TU Dortmund)

Description

System-theoretic dissipativity notions introduced by Jan C. Willems play a fundamental role in the analysis of optimal control problems. They enable the understanding of infinite-horizon asymptotics and turnpike properties. This talk introduces a dissipative formulation for training deep Residual Neural Networks (ResNets) in classification problems. To this end, we formulate the training of ResNets with a constant width as an optimal control problem and investigate its dissipativity properties when introducing a stage cost based on a variant of the cross entropy loss function, the classic loss function for classification tasks.

We illustrate the dissipative formulation by training on the MNIST dataset, which exhibits the turnpike phenomenon: the data remains unchanged throughout several layers. These layers can then be removed without changing the transformation learned by the NN. This technique can be used to obtain shallow neural networks for a given classification task with simplified hyperparameter tuning.

Authors

Jens Püttschneider (TU Dortmund) Timm Faulwasser (Upon submission)

Presentation materials

There are no materials yet.