Abstract:
|
Functional regression has played an essential role in inferring dependence structures among high-dimensional data associated with interest classes, such as disease status. Although functional regression offers a great advantage, the development of this model is quite limited due to its complexity and high dimensionality. The functional regressions were developed under the additive multi-functional regression assuming each function is independent of other functions. In this talk, we propose a sparse functional fused regression for correlated functional regression to solve the limitation of current multi-functional procedures. We develop our sparse functional fused regression under the Bayesian hierarchical model framework. Our approach can simultaneously conduct variable selection in nonparametric regression to select promising subsets. We demonstrate the advantage of our sparse functional fused regression using molecular profiling sequence data obtained from surface-enhanced Raman spectroscopy in bio-photonics to study drug dose level-response in brain cancer cells.
|