Apprentissage dans les réseaux récurrents pour la modélisation mécanique et étude de leurs interactions avec l'environnement

Abstract : Recurrent neural networks, which are inspired from biological principles, are used to model complex dynamic behaviours and to reproduce — to learn — such behaviours. The adaptive properties of these nets can be applied to physical modelling networks dedicated to the simulation of musical instruments. These physical modelling nets contain inertia, stiffness and damping parameters that one wishes to set automatically in order to reproduce a given mechanical behaviour ; this is possible thanks to recurrent neural algorithms. Algorithms for the paramater adaptation of physical models are developped and simulated. Original algorithms are also developed, based on mechanical principles.
Complete list of metadatas

https://tel.archives-ouvertes.fr/tel-00345820
Contributor : Thèses Imag <>
Submitted on : Wednesday, December 10, 2008 - 10:25:18 AM
Last modification on : Thursday, December 27, 2018 - 9:45:40 AM
Long-term archiving on : Saturday, November 26, 2016 - 3:12:19 AM

Identifiers

  • HAL Id : tel-00345820, version 1

Citation

Nicolas Szilas. Apprentissage dans les réseaux récurrents pour la modélisation mécanique et étude de leurs interactions avec l'environnement. Modélisation et simulation. Institut National Polytechnique de Grenoble - INPG, 1995. Français. ⟨tel-00345820⟩

Share

Metrics

Record views

1014

Files downloads

1187