Deniz Eroğlu (Imperial College London, Kadir Has University) – Data-Driven Recovery of Network Dynamics and Detection of Critical Transitions

Understanding and predicting critical transitions in complex systems—ranging from neural circuits to climate subsystems—requires the recovery of both their underlying dynamics and network structure directly from noisy and limited observations. In this talk, I will present a data-driven framework for reconstructing the governing equations and interaction topology of weakly coupled chaotic networks, combining ideas from model reduction and system identification. A key insight is to leverage stochastic fluctuations, typically considered as noise, as informative signals encoding the hidden network structure. These signatures allow us to infer effective models that combine local dynamics with statistically estimated coupling rules, enabling the prediction of critical regime shifts [1]. Under suitable assumptions, we further refine the approach to recover exact dynamics, and validate it using both synthetic data inspired by cortical circuits and experimental recordings from the mouse neocortex [2]. The method is robust to short time series and sparse sampling, making it applicable in practical settings where full observability is rarely achievable. I will conclude with a discussion of open challenges in the reconstruction of dynamical networks from partial data, and outline how incorporating concepts from normal form theory and synchronization dynamics helps overcome current limitations [3].

References:

[1] D. Eroglu, M. Tanzi, S. van Strien, T. Pereira, Phys. Rev. X 10, 021047 (2020).

[2] I. Topal, D. Eroglu, Phys. Rev. Lett. 130, 117401 (2023).

[3] E. Nijholt, J.L. Ocampo-Espindola, D. Eroglu, I.Z. Kiss, T. Pereira, Nat. Commun. 13, 4849 (2022).