The widespread adoption of artificial neural networks for hologram synthesis can be attributed to their ability to improve image quality and reduce computational costs. In this study, we propose an alternative use of artificial neural networks to improve the optical efficiency of complex field encoding. The neural encoding significantly enhances the efficiency of amplitude-only SLMs, resulting in 2.4-fold optical efficiency enhancement with negligible image quality degradation compared to the Burch encoding method. Notably, the experimental results demonstrate that the neural encoding method has even higher image quality, providing an approximately 2.5 dB enhancement in a peak signal-to-noise ratio. The neural encoding method offers promise in mitigating a fundamental challenge of conventional amplitude-only holograms, namely its low efficiency.