Quaternion recurrent neural network with real-time recurrent learning and maximum correntropy criterion
This addresses robust motion prediction for lung cancer radiotherapy, but it is incremental as it combines existing techniques in a novel quaternion framework.
The paper tackled robust real-time processing of 3D/4D data with outliers by developing a quaternion recurrent neural network using real-time recurrent learning and maximum correntropy criterion, showing improved outlier sensitivity in simulations for lung cancer radiotherapy motion prediction.
We develop a robust quaternion recurrent neural network (QRNN) for real-time processing of 3D and 4D data with outliers. This is achieved by combining the real-time recurrent learning (RTRL) algorithm and the maximum correntropy criterion (MCC) as a loss function. While both the mean square error and maximum correntropy criterion are viable cost functions, it is shown that the non-quadratic maximum correntropy loss function is less sensitive to outliers, making it suitable for applications with multidimensional noisy or uncertain data. Both algorithms are derived based on the novel generalised HR (GHR) calculus, which allows for the differentiation of real functions of quaternion variables and offers the product and chain rules, thus enabling elegant and compact derivations. Simulation results in the context of motion prediction of chest internal markers for lung cancer radiotherapy, which includes regular and irregular breathing sequences, support the analysis.