The utilization of data from various materials databases is crucial in the development of accurate ML models for materials discovery. However, it has been observed that a considerable portion of the entries in these databases originate from experimental data, which inevitably limits the structural diversity of the materials contained therein. To address this issue, our research aims to unveil previously undiscovered crystal structures by employing a coarse-graining approach that characterizes crystal structures based on their symmetries. This proposal is to investigate on a large-scale screening of crystal structures predicted to be stable through their symmetries, as detailed in our published work (https://doi.org/10.1126/sciadv.abn4117). In practice we will primarily run pre-trained ML model inference for energy predictions using our own pytorch/JAX Python code.