Dynamic causal modelling in probabilistic programming languages DOI Creative Commons
Nina Baldy, Marmaduke Woodman, Viktor Jirsa

и другие.

Journal of The Royal Society Interface, Год журнала: 2025, Номер 22(227)

Опубликована: Июнь 1, 2025

Understanding the intricate dynamics of brain activities necessitates models that incorporate causality and nonlinearity. Dynamic causal modelling (DCM) presents a statistical framework embraces relationships among regions their responses to experimental manipulations, such as stimulation. In this study, we perform Bayesian inference on neurobiologically plausible generative model simulates event-related potentials observed in magneto/encephalography data. This translates into probabilistic latent states system driven by input stimuli, described set nonlinear ordinary differential equations (ODEs) potentially correlated parameters. We provide guideline for reliable presence multimodality, which arises from parameter degeneracy, ultimately enhancing predictive accuracy neural dynamics. Solutions include optimizing hyperparameters, leveraging initialization with prior information employing weighted stacking based accuracy. Moreover, implement conduct comprehensive comparison several programming languages streamline process benchmark efficiency. Our investigation shows inversion DCM extends beyond variational approximation frameworks, demonstrating effectiveness gradient-based Markov chain Monte Carlo methods. illustrate efficiency posterior estimation using self-tuning variant Hamiltonian automatic Laplace approximation, effectively addressing degeneracy challenges. technical endeavour holds potential advance state-space ODE models, contribute neuroscience research applications neuroimaging through DCM.

Язык: Английский

Dynamic causal modelling in probabilistic programming languages DOI Creative Commons
Nina Baldy, Marmaduke Woodman, Viktor Jirsa

и другие.

Journal of The Royal Society Interface, Год журнала: 2025, Номер 22(227)

Опубликована: Июнь 1, 2025

Understanding the intricate dynamics of brain activities necessitates models that incorporate causality and nonlinearity. Dynamic causal modelling (DCM) presents a statistical framework embraces relationships among regions their responses to experimental manipulations, such as stimulation. In this study, we perform Bayesian inference on neurobiologically plausible generative model simulates event-related potentials observed in magneto/encephalography data. This translates into probabilistic latent states system driven by input stimuli, described set nonlinear ordinary differential equations (ODEs) potentially correlated parameters. We provide guideline for reliable presence multimodality, which arises from parameter degeneracy, ultimately enhancing predictive accuracy neural dynamics. Solutions include optimizing hyperparameters, leveraging initialization with prior information employing weighted stacking based accuracy. Moreover, implement conduct comprehensive comparison several programming languages streamline process benchmark efficiency. Our investigation shows inversion DCM extends beyond variational approximation frameworks, demonstrating effectiveness gradient-based Markov chain Monte Carlo methods. illustrate efficiency posterior estimation using self-tuning variant Hamiltonian automatic Laplace approximation, effectively addressing degeneracy challenges. technical endeavour holds potential advance state-space ODE models, contribute neuroscience research applications neuroimaging through DCM.

Язык: Английский

Процитировано

0