Projection rule for complex-valued associative memory with large constant terms

Abstract

Complex-valued Associative Memory (CAM) has an inherent property of rotation invariance. Rotation invariance produces many undesirable stable states and reduces the noise robustness of CAM. Constant terms may remove rotation invariance, but if the constant terms are too small, rotation invariance does not vanish. In this paper, we eliminate rotation invariance by introducing large constant terms to complex-valued neurons. We have to make constant terms sufficiently large to improve the noise robustness. We introduce a parameter to control the amplitudes of constant terms into projection rule. The large constant terms are proved to be effective by our computer simulations.