Semiparametric and Graphical Models

The figure above, taken from Na et al., 2020, displays the glass brains depicting the estimated differential network between the schizophrenia and control groups based on an fMRI dataset.

Semiparametric models allow for flexible modeling of data without making rigid assumptions the parametric models make, while at the same time allowing for tractable estimation without suffering from the curse of dimensionality that obsesses fully nonparametric models. I have conducted research on semiparametric index models (Na and Kolar, 2021, Na et al., 2019) and semiparametric matrix completion problems (Na et al., 2020).

Additionally, I have worked on graphical models, which are utilized to capture complex relationships among observed variables and have applications in genetics, neuroscience, and computational biology. See Na et al., 2020 for the estimation of differential networks under the presence of latent factors.