The human brain shows complex connectivity patterns at multiple spatial and temporal scales which are fundamental to the emergence of cognition and intelligence. My primary research goal is to develop innovative analysis tools and computational models (i) to capture and understand the patterns of structural and functional connectivity in the human brain, (ii) to differentiate these patterns in health and disease, and (iii) to harness this understanding to build intelligent systems.
I utilize a dynamical systems perspective and visualize the brain as an adaptive, complex network (graph) of sub-systems. Analysis tools I develop are built upon the ideas derived from mathematics, physics, statistics, and computer science.
Here is a brief description of my research projects:
How does brain activity reflect shifting cognitive states—and how can we model this variability using machine learning in a meaningful, interpretable way? Our ongoing work explores this question by applying deep neural networks (1D-CNN and BiLSTM) to fMRI data from cognitive tasks. Rather than focusing solely on classification accuracy, we emphasize model interpretability and behavioral relevance.
Our recent paper in Scientific Reports shows that (i) individual differences in task performance systematically influence prediction accuracy and (ii) visual, attention, and control networks distinguish cognitive state dynamics.
How does intelligence emerge from the underlying structural connectome? We build personalized brain network models to be able to assess the relationship between brain’s structural connectivity and emergent functional dynamics, and gain predictive knowledge of individual cognitive abilities.
Our review on personalized brain network models: Current Opinion in Neurobiology, 52, 42-47 (2018).
Understanding cognitive relevance of brain connectivity [Science Advances, 5 (4), eaau8535 (2019)].
Predicting individual performance on language tasks [Plos Computational Biology, 14 (10), e1006487 (2018)].
Mounting evidence supports that modular organization is characteristic to brain's functional dynamics. How and why do these modules dynamically reconfigure? This line of work focuses on capturing the rapid reconfigurations in brain network modules. We leverage several network science tools (and also add our tweaks for some improvements here and there!).
In a series of papers we show how brain modularity, and brain flexibility (likelihood of module reconfigurations), relates to adaptability, robustness, and even healthy functioning of the brain.
Brain is always active, even when we are resting. What's the meaning of spontaneous brain activity and what does it encode?
Spontaneous activity in resting state EEG data can be captured by the wide-spread spatiotemporal cascades of high-amplitude bursts or (macroscopic) neural avalanches.
We observed that the spatial avalanche distribution (which we call 'normalized engagement') changes with changing stimuli and can be predictive of the stimulus-driven, individualistic cognitive responses. [NeuroImage, 241, 118425 (2021)].
I am exploring the application of capacitance-based, ankle-worn devices with ML based techniques in detecting cortical arousal, sleep-wake states, and sleep related disorders (in collaboration with Tanzen Medical Inc and D-prime LLC).
Before making a transition into neuroscience, I studied optoelectronic devices and obsevred interesting emergent properties, both by investigating an individual device as well as a network of devices.
Network symmetries predict emergent patterns of synchronization
With Prof. Rajarshi Roy, University of Maryland, College Park.
Symmetries in a network’s structure can help predict the emergent patterns of synchronized clusters in a network of coupled oscillators. In a globally coupled system of opto-electronic chaotic oscillators, we showed that the symmetries and sub-symmetries of the network predict the formation of synchrony and partial synchrony states including chimera states, where system evolves into separate domains of synchronized and desynchronized populations. We further showed that the master stability equation can be significantly simplified by usind symmetry analysis with a group theoretical approach, and stability of chimera and cluster states can be successfully predicted [Chaos, 26 (9), 094801 (2016)].
PhD dissertation: Electrical and optical investigations of the condensed matter physics of junction diodes under charge carrier injection
With Dr. Shouvik Datta, Indian Institute of Science education and Research (IISER), Pune, India.
Lasers are complex systems where coherent light emission is achieved as the system is directed (by design) to enhance particular interactions. A diode laser is a special case where fermionic populations achieve a state analogous to population inversion by charge carrier injection. We demonstrated that the conventional electrostatic description of a junction diode breaks down with increasing injected charge density. We captured this departure as coupled, frequency dependent optical (modulated light emission) and electrical (capacitance) properties of the device. We observed the occurrence of inductive like ‘negative capacitance’ simultaneously with the onset of low frequency dependent modulated light output, and explained this electrical and optical correlation through the participation of defect energy levels in light emission process [J. Appl. Phys. 110, 114509 (2011)]. We further demonstrated that increasing quantum confinement of charge carriers (quantum wells to quantum dots) affects the overall light emission process and identified electrical signature of excitonic Mott transition [Appl. Phys. Lett. 102, 053508 (2013); J. Appl. Phys. 120, 144304 (2016)]. We established that to construct an efficient device, one cannot completely rely on characterizing individual constituent material layers but the dynamics of the entire combined system needs to be understood [Appl. Phys. Lett. 105, 123503 (2014)].