Author ORCID Identifier

0009-0002-8894-1990

Author Linked-In Account

https://www.linkedin.com/in/dr-hariharan-srinivasan-9bb796183/

Biosketch

HariharanSrinivasan received his B.Sc. degree in Physics from SCSVMV University, Kanchipuram, Tamil Nadu, India, in 2019, securing a CGPA of 9.13/10. He obtained his M.Sc. degree in Physics from SASTRA Deemed to be University, Thanjavur, Tamil Nadu, India, in 2021 with a CGPA of 8.68/10. He pursued his doctoral studies in Physics at SASTRA Deemed to be University from 2022 to 2025.

His research interests encompass nonlinear dynamics, computational neuroscience, complex networks, stochastic processes, reservoir computing, astrophysics and machine learning. His doctoral research focuses on investigating noise-induced extreme events in neuronal oscillators and networks and developing machine-learning frameworks for forecasting complex dynamical systems. He has also worked on the development of clustered Echo State Network architectures to improve the prediction of nonlinear time series and extreme events.

During his doctoral research, he has authored five research articles published in peer-reviewed international journals. His research combines concepts from nonlinear science, physics, network theory, and machine learning to understand and predict emergent behavior in complex systems. His broader academic goal is to advance the understanding of complex dynamical systems by investigating the mechanisms underlying natural phenomena and their prediction using machine learning.

Date of Award

19-2-2026

Document Type

Thesis

School

School of Electrical & Electroncis Engineering

Programme

Ph.D.-Doctoral of Philosophy

First Advisor

Dr.R.Suresh

Keywords

Extreme events, Neuronal Oscillators, Stochastic systems, Complex networks, Noise-induced Transitions, Echo state network, Nonlinear Dynamics

Abstract

This doctoral dissertation comprehensively investigates the underexplored phenomenon of noise-induced extreme events (EE). The word “extreme” is accompanied by an occurring “event” when the deviation is extreme or higher than that of regular occurrences. These extreme occurrences are rare, abrupt, sudden, and irregular, often causing a profound impact on the system and its surroundings. Tsunami, earthquakes, solar flares, and tornadoes are such events that do not occur often but still significantly cause damage to mankind. This thesis particularly focuses on EE in neuronal systems where sudden synchronization can trigger seizures, tremors, and strokes which serve as classic examples of such sporadic events.

Neurons are inherently nonlinear entities that exhibit spiking and resting. But the state of spiking is decided by several factors such as the strength of external inputs, internal processes, and fluctuating signals from neighboring neurons due to complex biological mechanisms. To understand these mechanisms, nonlinear dynamical models have been employed to replicate neuronal behavior in both isolated and collective settings. In such models, EE have been studied in a deterministic setting, neglecting the realistic influence of noise on neuronal systems. Hence, this thesis centralizes its investigation on exploring the influence of noise in inducing EE.

The first part of the thesis focuses on understanding the behavior of neurons in single units with memristive Hindmarsh-Rose and FitzHugh-Nagumo models, in which the role of noise in inducing EE is thoroughly investigated with numerical and analytical techniques, particularly focusing on their energy aspect. The study is further extended to all-to-all coupled neuronal networks, where non-identical noise inputs mimic realistic biological scenarios, thereby elucidating the causal mechanisms of EE at both microscopic and macroscopic scales. Further, complex structures such as the multiplex network are investigated to decipher the role of noise and coupling when two layers interact with neurons receiving heterogeneous noise sources.

Since these sporadic events occur without precursors or early warning signals, predicting them poses a significant challenge, especially in multivariate time series. To tackle this, the thesis offers a novel prediction methodology by modifying the conventional Echo State Network (ESN) into a clustered ESN that mimics brain modularity, enabling efficient prediction of EE in a multivariate time series through various network topologies. The findings of this thesis address the underexplored phenomenon of noise-induced EE in neuronal models of both isolated and network configurations by identifying the dynamical mechanisms influenced by noise, while also introducing a novel machine learning algorithm to predict multivariate EE, a challenge not previously addressed.

Download

Share

COinS