Dr Sandip George

Biography

My work focuses on using the framework of nonlinear dynamics and machine learning to understand data from a range of different fields. As part of my research I have worked on data from astronomy, ecology, psychiatry and critical care medicine.

I completed my PhD in Physics from IISER Pune, in India,  working on nonlinear time series analysis. My work focused on technique development and using these to understand the dynamics of the light variation in stars. Subsequently, I moved to the University Medical Center Groningen (UMCG), the University of Groningen, in the Netherlands, where I was working on understanding and predicting transitions to depression. This was mostly done by studying the variations in the nonlinear dynamics of mood, movement and cardiac dynamics during or prior to a depressive episode. Prior to my appointment at Aberdeen, I was a research fellow at the CHIMERA hub at the University College London, where I worked on understanding and predicting how patients in the intensive care unit evolve. The work focused on leveraging machine learning, dynamical systems and modelling to study specific problems in critical care.

I am highly interested in problems in nonlinear time series analysis, either from the point of view of understanding dynamical systems, or as applied to understand specific real-world contexts. Please send me an email if you'd like to explore any of topics listed above further, or in the context of your own research.

External Memberships

Honorary Lecturer- University College London

Academic editor- PLoS One

Fellow- Royal Astronomical Society

Research Overview

My work broadly uses time series analysis and dynamical systems to understand problems from a number of fields. The techniques that I use are mainly derived from the field called nonlinear time series analysis. A lot of this is based on the idea that the full dynamics of a system of coupled differential equations can be reconstructed from the dynamics of a single observable. The topological properties of this reconstructed phase space is the same as the properties of the original phase space. Quantifying the properties of this space for real-world time series and relating them back to the physical (or biological) properties of the system from which they were derived is a theme that recurs in my work.

One of the other techniques that I use is the Fourier transforms of higher order moments, such as the bispectrum or trispectrum. This is an underutilized technique in the context of dynamical systems, but lead to powerful insights about the nature of periodicities present in such systems.

Finally, I also work on combining ideas from nonlinear dynamics and machine learning for modelling and classification of real world systems. In the context of the former, methods from scientific machine learning are employed to study partially known systems of differential equations. In the latter, I use techniques from nonlinear time series analysis as features to classify standard systems, with the understanding that these measure aspects of the time series that are not captured using standard quantifiers such as moments of the amplitude distribution.

Current Research

Astronomy: Nonlinear dynamics of close binaries

My research in astronomy revolves around understanding the dynamics of variable stars. In particular I am interested in the dynamics of close eclipsing binary stars, i.e. binary stars that can exchange mass and/or energy with each other. While the light variation due to eclipses is periodic, the mass exchange results in variations that are irregular. The properties of the reconstructed space of these stars are closely related to astrophysical properties of these stars. In particular, the morphological classification of these stars is closely related to their recurrence properties[1]. Secondly, in a subset of these stars, called overcontact binaries, the extend of closeness between the component stars (measured using a fill-out factor) is shown to be closely related to their nonlinear properties[2].

[1] George, S. V., Misra, R., & Ambika, G. (2019). Classification of close binary stars using recurrence networks. Chaos: An Interdisciplinary Journal of Nonlinear Science, 29(11).

[2] George, S. V., Misra, R., & Ambika, G. (2020). Fractal measures and nonlinear dynamics of overcontact binaries. Communications in Nonlinear Science and Numerical Simulation, 80, 104988.

Critical Care Medicine: Discontinuing mechanical ventilation

Predicting patient deterioration is a question of critical importance in healthcare, but in particular in the intensive care unit. Making these predictions using data that is routinely measured in the ICU has been a research thrust in recent years. A specific question of interest in this context is to predict when a patient is likely to breathe without the help of mechanical ventillation. Removal of the breathing tube is termed extubation. Recently we attempted to check if signatures seen in dynamical systems, prior to critical transitions, can be seen post extubation in patients who fail extubation[1].

[1] Khalil, L., George, S. V., Ray, S., & Arridge, S. (2022, September 7). Transitions in intensive care: Investigating critical slowing down post extubation. https://doi.org/10.17605/OSF.IO/3DWTA

Psychiatry: Physical activity, heart rate and their relation to depression

While transitions to depression may appear gradually over the course of several weeks, they may also occur abruptly, leading to sudden transitions. Predicting the latter could lead to timely interventions that could prevent the episode or mitigate its effects. Methods based on ecological momentary assessment have been used commonly to make such predictions. However such methods require higher patient burden for measurement than physical activity. In a recurrence quantification analysis of 21 currently depressed and 25 non-depressed individuals, we identified that depressed people showed lower duration and diversity of recurrent physical activities than individuals without depression [1]. A second measurement we used to predict depressive transitions was the ECG. The complexity of cardiac dynamics is thought to reduce considerably during depression. However, it is not understood if this reduction precedes depression or occurs as a result of it. In a study of 28 individuals tapering antidepressants, 14 of whom experienced an episode of depression and 14 who did not, we observed a reduced complexity (measured using multiscale entropy) of interbeat intervals in the weeks leading up to the episode, suggesting that this reduction precedes depressive episodes [2].

[1] George, S. V., Kunkels, Y. K., Booij, S., & Wichers, M. (2021). Uncovering complexity details in actigraphy patterns to differentiate the depressed from the non-depressed. Scientific Reports, 11(1), 13447.

[2] George, S. V., Kunkels, Y. K., Smit, A., Wichers, M., Snippe, E., van Roon, A. M., & Riese, H. (2023). Predicting recurrence of depression using cardiac complexity in individuals tapering antidepressants. Translational Psychiatry, 13(1), 182.

Past Research

Did Betelgeuse undergo a critical transition?

In late 2019, the variable star Betelgeuse underwent a dimming episode that surpassed any previous reductions in its brightness. A number of hypotheses were raised for this reduction, the most popular of which was a dust cloud. An analysis of the light variations leading up to this episode suggested that the autocorrelation and the variance increased significantly, apart from recurrence based measures. Such increases are characteristic of critical transitions in dynamical systems. Hence we hypothesized that the reasons behind the dimming is internal to the star, and could have resulted from a change in the dynamics responsible for its variation, for instance pulsations[1,2]. Recent results have shown that since the dimming, the star has started showing a new dominant pulsational mode (see here and here).

[1] George, S. V., Kachhara, S., Misra, R., & Ambika, G. (2020). Early warning signals indicate a critical transition in Betelgeuse. Astronomy & Astrophysics, 640, L21.

[2] Kachhara, S., George, S. V., Misra, R., & Ambika, G. (2023). Evidence for dynamical changes in Betelgeuse using multi-wavelength data. In The Sixteenth Marcel Grossmann Meeting on Recent Developments in Theoretical and Experimental General Relativity, Astrophysics and Relativistic Field Theories: Proceedings of the MG16 Meeting on General Relativity Online; 5–10 July 2021 (pp. 3485-3493).

Collaborations

A non-exhaustive list of people who I have worked with and/or continue to work with.

Prof G. Ambika, Indian Institute of Science Education and Research, Thiruvananthapuram, India

Prof. Ranjeev Misra, Inter-University Center for Astronomy and Astrophysics, Pune, India

Prof. Marieke Wichers, University Medical Center Groningen, Groningen, The Netherlands

Dr. Harriëtte Riese, University Medical Center Groningen, Groningen, The Netherlands

Dr. Arie van Roon, University Medical Center Groningen, Groningen, The Netherlands

Dr. Sneha Kacchara, Northwestern University, Evanston, USA

Prof. Simon Arridge, University College London, London

Dr. Samiran Ray, Great Ormond Street Hospital, London

Funding and Grants

Royal Society Research Grants (2024 Round 1)- 'Detecting cardiac pathology from ECG using dynamical models and data' (£18,373.14)

Teaching Responsibilities

MX 4085- Nonlinear Dynamics and Chaos I (First half session 2023-'24)