Article ID: 2026MAP0002
In a few decades, many researchers have attempted to develop neural networks using Clifford algebra, also referred to as geometric algebra. Algebra of split-quaternions is a four-dimensional Clifford algebra, like that of quaternions, and split-quaternion-valued Hopfield networks (SQVHNs) have been proposed as extensions of Hopfield networks. Projection rule is a practical learning algorithm. Exact stability conditions are necessary for effective implementation of projection rule, referred to as noise robust projection rule. Recently, the exact stability conditions for SQVHNs were determined. In this paper, we propose the noise robust projection rule for SQVHNs. We conduct computer simulations and show that the proposed projection rule provides better noise tolerance than the conventional one.