The scientific method is based on the observation of phenomena, building hypotheses and making measurements to compare theoretical expectation with observation. Is it possible to follow the same process while deducing the underlying model directly from the data without a specific prior expectation? With recent machine learning techniques the answer might be yes and the SINDy framework might be one possible approach to tackle the problem. Sparse Identification of Nonlinear Dynamics (SINDy) is based on machine learning techniques, it gets empirical data as an input and outputs the differential equation(s) that most likely have produced the observed time course of observables of physical interest, by means of an optimization approach. A Thesis work is here proposed to test the SINDy framework, and discover its limits in the context of complex networks dynamics. Specifically, we will explore distinct dynamical processes, such as synchronization via Kuramoto-like models, on the top of a network with varying topology, encoding increasing levels of complexity observed in empirical biophysical systems (such as heterogeneous connectivity, modular and hierarchical structure, spatial embedding). The goal is to understand to which extent machine-assisted approaches like SINDy can help to better understand the structure and the dynamics of empirical complex systems.
Il metodo scientifico si basa sull'osservazione dei fenomeni, la formulazione di ipotesi, e il fare misurazioni per mettere a confronto le aspettative teoriche con i dati osservati. È possibile seguire lo stesso processo mentre si deduce il modello sottostante direttamente dai dati senza una specifica aspettazione a priori? Con le tecniche recenti di machine learning la risposta potrebbe essere sì e il framework SINDy potrebbe essere un possibile approccio per affrontare il problema. Sparse Identification of Nonlinear Dynamics (SINDy) si basa su tecniche di machine learning, prende dati empirici come input e restituisce le equazioni differenziali che con maggior probabilità hanno prodotto le timeseries di osservabili di interesse fisico, attraverso un approccio di ottimizzazione. Viene proposto qui un lavoro di tesi per testare il framework SINDy, e scoprire i suoi limiti nel contesto di dinamiche di reti complessi. In particolare esploreremo diversi tipi di dinamica come la sincronizzazione attraverso modelli basati su quello di Kuramoto, su reti con diverse topologie che racchiudono diversi livelli di complessità osservata nei sistemici empirici biofisici (come la connettività eterogenea, modulare, a struttura gerarchica, o spaziale). L'obiettivo è capire fino a che punto gli approcci di machine-assisted come SINDy possono aiutare per capire meglio la struttura e la dinamica di sistemi complessi empirici.
Scoprire le leggi del moto dalle dinamiche collettive su reti con il framework SINDy
VALSECCHI, ANDREA
2024/2025
Abstract
The scientific method is based on the observation of phenomena, building hypotheses and making measurements to compare theoretical expectation with observation. Is it possible to follow the same process while deducing the underlying model directly from the data without a specific prior expectation? With recent machine learning techniques the answer might be yes and the SINDy framework might be one possible approach to tackle the problem. Sparse Identification of Nonlinear Dynamics (SINDy) is based on machine learning techniques, it gets empirical data as an input and outputs the differential equation(s) that most likely have produced the observed time course of observables of physical interest, by means of an optimization approach. A Thesis work is here proposed to test the SINDy framework, and discover its limits in the context of complex networks dynamics. Specifically, we will explore distinct dynamical processes, such as synchronization via Kuramoto-like models, on the top of a network with varying topology, encoding increasing levels of complexity observed in empirical biophysical systems (such as heterogeneous connectivity, modular and hierarchical structure, spatial embedding). The goal is to understand to which extent machine-assisted approaches like SINDy can help to better understand the structure and the dynamics of empirical complex systems.File | Dimensione | Formato | |
---|---|---|---|
Andrea_Valsecchi.pdf
accesso aperto
Dimensione
4.93 MB
Formato
Adobe PDF
|
4.93 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/84644