Unseen young planets can be characterized by analyzing dust emissions from protoplanetary disks. The mass of embedded planets is inferred through numerical simulations, empirical relations, or deep learning models. In this study, we employ Variational Autoencoders (VAEs) to infer planetary parameters from simulated protoplanetary disk images while quantifying the uncertainties in the predictions. Our approach provides a robust framework for parameter estimation and uncertainty quantification, enhancing the reliability of planet characterization from disk observations.