Solution of Inverse Problems Using Multilayer Quaternion Neural Networks

Neural network models extended to higher-dimensional numbers have been studied in recent years. In particular, quaternions have an advantage with respect to the expression of rotation in a three-dimensional space. On the other hand, the problem that estimates the cause of an observed result is called an inverse problem, whose solutions have been studied in various engineering fields. In this study, the quaternion network inversion is proposed as a neural network solution of the inverse problem extended to the quaternion. The simulation of the inverse mapping problem is examined to show the effectiveness of the proposed method.