|
Abstract:
|
Advances in technology and science have spurred an increasing accumulation of complex data sources, leading modern researchers into a diverse present-day data revolution. The new tenor of data collection has created a two-part problem: historical models are often not malleable enough to accommodate data diversity, all the while connections among multiple data sources are being identified as more promising expressions of scientific processes. These new challenges are requiring data scientists to abandon model-based algorithms for data-driven techniques which permit a multi-modal approach to data analysis. A brief discussion of the theoretical underpinnings and computational challenges will be presented to compare decomposition analyses and deep learning methods under this framework. Then, a novel, hybrid approach is extended to the algorithmic learning that permits the combination of multiple sources of information, while retaining interpretability, generalizability, and the capability of downstream analyses. This method has a direct application to imaging genetics, where the aim is to fuse various modalities of neuroimaging with genetic markers to elucidate the progression of neur
|
