Background <p>Conventional acoustic materials exhibit limited effectiveness in attenuating low-frequency noise, motivating the development of subwavelength resonant acoustic metasurfaces.</p> Methods <p>This study proposes a hyperbolic-neck Helmholtz resonator (HNHR) and systematically investigates the sound-absorption behavior of low-frequency acoustic metasurfaces formed by coupling multiple HNHR units using theoretical modeling, finite-element simulations, and experimental measurements. An autoencoder-like neural network (ALNN) is developed for the inverse design of structural parameters, enabling rapid determination of geometries that meet target absorption requirements.</p> Results <p>The results show that the predicted responses of network-designed structures at different scales exhibit excellent agreement with both simulations and experimental data within the target frequency band, achieving the desired absorption performance.</p> Conclusions <p>These findings demonstrate the feasibility and effectiveness of the proposed method for rapid inverse design of low-frequency acoustic metasurfaces.</p>