Document type
Bachelor thesisPublication date
Publication license
Please use this identifier to cite or link to this item: https://hdl.handle.net/2445/187661
Simulation of physical systems with variants of the Interaction Network
Journal Title
Authors
Director/Tutor
Journal ISSN
Volume Title
Related resource
Abstract
[en] The aim of this thesis is to primarily learn to predict trajectories of physical systems by using Machine Learning. We have used the Interaction Network as a base model and introduced tweaks to its structure in the framework of Graph Networks in order to deal with systems without interaction, with force-based interactions and systems governed by the Vicsek model. We have been able to replicate very well systems without interaction and systems governed by a Vicsek model of infinite reach. However the results with force-based systems are mediocre because they need more trainable parameters and training data. The results for the Vicsek model with a finite radius of reach are the worst but we have learned the necessity and the methodology
of introducing attention to the base model to deal with this class of problem.
Description
Treballs finals del Màster de Fonaments de Ciència de Dades, Facultat de matemàtiques, Universitat de Barcelona. Curs: 2020-2021. Tutor: Albert Díaz Guilera i Oriol Pujol Vila
Subject (English)
Citation
Citation
CÓRDOBA MENESES, Alfons. Simulation of physical systems with variants of the Interaction Network. [consulted: 8 of June of 2026]. Available at: https://hdl.handle.net/2445/187661