A1296
Title: Inference on the significance of modalities in multimodal generalized linear models
Authors: Quefeng Li - University of North Carolina - Chapel Hill (United States) [presenting]
Wanting Jin - UNC (United States)
Abstract: Multimodal statistical models have gained much attention in recent years, yet there is a lack of rigorous statistical inference tools for inferring the significance of a single modality within a multimodal model. This inference problem is particularly challenging in high-dimensional multimodal models. In the context of high-dimensional multimodal generalized linear models, a novel entropy-based metric called the expected relative entropy (ERE) is proposed to quantify the information gain of one modality in addition to all other modalities in the model. A deviance-based statistic is then proposed to estimate the ERE. The deviance-based statistic is proven consistent with the ERE, and its asymptotic distribution is derived, which enables the calculation of confidence intervals and p-values to assess the significance of the ERE for a given modality. The empirical performance of the proposed inference tool is numerically evaluated on various high-dimensional multimodal generalized linear models, and its good performance is demonstrated in these models. The method is also applied to a multimodal neuroimaging dataset to demonstrate its capability to infer the significance of imaging modalities, which is crucial for neuroscience studies.
