Cognitive Sciences and Psychology
Title : | Fully bayesian estimation of virtual brain parameters |
Area of research : | Cognitive Sciences and Psychology |
Focus area : | Computational Neuroscience |
Principal Investigator : | Dr. Jayant Jha, Indian Statistical Institute, West Bengal |
Timeline Start Year : | 2022 |
Timeline End Year : | 2024 |
Contact info : | jayantjha@gmail.com |
Equipments : | Laptop |
Details
Executive Summary : | The study aims to propose novel methodologies to efficiently and accurately quantify the uncertainty associated with model parameters based on empirical SEEG and fMRI data from subjects with partial epilepsy and healthy subjects. The fully Bayesian methodology can provide a reference for analyzing empirical datasets based on other high-dimensional and complex state-space models. The research tasks include developing statistical methodologies based on self-tuning Hamiltonian Monte Carlo (HMC) algorithm to infer the parameters of the whole-brain network model given by the stochastic differential equations based model proposed in Montbrio et al., 2015, reducing the false positive rate in the identification of epileptogenic zones based on empirical SEEG data and whole-brain network model.
The study also aims to develop methodologies to exploit the expressive power of deep learning to serve Bayesian inference in the spirit of simulation-based methods. The core of the methodology is only requiring forward-simulations from the computer model, rather than model-specific analytic calculation or exact evaluation of likelihood function using MCMC algorithms. The strategy using SBI requires fast forward-simulations, sufficient memory, system infrastructures, and informative data features extracted from the forward-model, offering a flexible approach to model prediction and validation in complex systems. |
Total Budget (INR): | 14,94,669 |
Organizations involved