Comparing Static and Dynamic Artificial Intelligences on their Ability to
Predict Patient Outcomes in the NACC Database 
   Lead Investigator:    Thomas Anastasio
   Institution      :     
   E-Mail           :     
   Proposal ID      :    1218



   Proposal Description:
   By far the most powerful Artificial Intelligences (AIs) in use today are deep (multi-layer) neural networks. Some initial attempts to extract knowledge from the NACC database have been made using deep (and shallow) networks but these have been static, in the sense that the networks had only feedforward connections and so could not process information in time. Recurrent networks are dynamic, because they have feedforward and feedback connections, and so can process information in time. They are also processing in a recurrent network is equivalent to a layer in a feedforward network. The NACC dataset has temporal information in the form of sequential patient visits, but this temporal information has not yet been accessed by AIs. The goal of this project is to use both static and dynamic neural networks of various types to extract the knowledge contained in the NACC database, and to     compare the networks by testing each ones ability to predict later health status from previous health information. We expect that recurrent networks will make more accurate predictions than feedforward networks because we assume that temporal information is of utmost importance. The results of this study will let us know for sure. The proposed work will have great impact because it will help us decide which AIs are the most powerful AIs for extraction of the knowledge contained in the NACC database.