The rapid advancements in additive manufacturing have facilitated the development of high-performance cellular solids, including truss-, shell-, and plate-lattices. Among these, plate-lattices are particularly notable for their exceptional mass-specific stiffness, approaching theoretical performance limits [1]. This study introduces a systematic approach for generating plate-lattices by sequentially placing plates within a unit cell, characterized by an anchor point, a normal vector, and a thickness, while satisfying periodic boundary conditions [2]. Using this method, we construct a dataset of over 90,000 plate-lattices and evaluate their anisotropic stiffness properties through finite element analysis. Our findings indicate that plate arrangement enables significant control over macroscopic mechanical properties without increasing material usage. Additionally, we establish a clear relationship between plate orientation and stiffness and compare the property spaces of plate-lattices with those of truss- and shell-lattices. To tackle the inverse design problem—determining optimal lattice structures for targeted mechanical properties—we develop a neural network incorporating LSTM and feedforward layers. The forward model undergoes supervised training, while the inverse model is trained with feedback from a pre-trained forward model. Through extensive hyperparameter optimization, we ensure the model’s ability to generalize effectively, enabling the efficient design of plate-lattices with tailored mechanical performance. This work advances inverse design methodologies, offering a powerful tool for optimizing the stiffest class of metamaterials.