Before the Crash: How Machines Learn to Predict Chaos

=========

Abstract

=========

Many systems evolve smoothly and then suddenly change — a market crashes, a climate pattern shifts, or a chaotic attractor disappears. Predicting such abrupt transitions is one of science’s hardest problems. Reservoir computing, a machine-learning approach, has shown impressive forecasting power, but what does the machine really learn? In this talk, I show how reservoir computing can anticipate dynamical crises in simple nonlinear systems such as the logistic and Gauss maps. By uncovering how the reservoir’s internal activity mirrors the underlying dynamics, we gain an intuitive picture of how machines learn to foresee dramatic change before it happens.

 

=================

About the Speaker

=================

Dr. Sarika Jalan is a Professor at the Indian Institute of Technology Indore, specialising in complex systems, network theory, and nonlinear dynamics. Her research explores synchronisation, coupled Kuramoto oscillators, multilayer and higher-order networks, machine learning, Chaos and financial systems.

 

Prior to joining IIT Indore, she held postdoctoral and research scientist positions at Max Planck Institute for Mathematics in the Sciences (Leipzig), the Max Planck Institute for the Physics of Complex Systems (Dresden), and the National University of Singapore (Singapore). She has also served as a visiting professor at TU Berlin (Germany), Institute for Basic Science, Daejeon (South Korea), Institute for Complex Systems, Florence (Italy), and Lobachevsky University, Nizhny Novgorod (Russia). Dr. Jalan is section Editor of Chaos, Solitons and Fractals. Prior to this, she held Editor-in-Chief position of the Journal of Computational Science (2021-2025), advisory editor at Chaos: An interdisciplinary journal on nonlinear science, Editor of EPJB and Nature Scientific reports. She was awarded prestigious SERB Power Fellowship (2022-2025) by Government of India.

 

Dr. Jalan is Board Member of the Network Science Society (USA) and has served on the executive committee of the Complex Systems Society (France). She is also an invited member of the Bernoulli Society Committee for Statistical Network Science. She received outstanding service award of 2025 from Netsci Society.