notesum.ai

Published at December 4

Hamiltonian-based neural networks for systems under nonholonomic constraints

physics.class-ph

cs.LG

Released Date: December 4, 2024

Authors: Ignacio Puiggros T.¹, A. Srikantha Phani

Aff.: ¹University of British Columbia

Arxiv: http://arxiv.org/pdf/2412.03018v1

Refer to caption

Parameter	$H$	$\bm{\lambda}$	A
Hidden layers	2	2	2
Neurons per layer	200	300	400
Activation function	$Tanh$	$Tanh$	$Tanh$
Learning rate	5e-5	5e-5*	5e-5*
Epochs	5e5
Training set	4.8e4 points
Validation set	3e3 points
Batches	48