Neural Network Prediction of Reduced Ion Mobility of Chemical Compound Based on Molecular Structure

We present a user-friendly hardware learning algorithm called the cascade error projection (CEP) that was developed at JPL and was equipped with a new input feature mapping technique. This new technique is based on Riemannian metric tensor to enhance the learning capability for predicting the reduced ion mobility based on the molecular structure. Our simulation results are reported and compared with the current state-of-the-art ADAPT tools developed by Pennsylvania State University. In addition, our approach is superior in our novel hardware implementation approach enabling a low power, low cost and miniaturized system for remote applications e.g., NASA mission.