Zhang, Y., Singh, S., Thanoon, D., Devarakota, P., Jin, L. and Tsvankin, I., 2023. Physics-guided unsupervised deep-learning seismic inversion with uncertainty quantification. Journal of Seismic Exploration, 32: 257-270. Data-driven seismic inversion techniques are often used for estimation of subsurface properties. Employing the acoustic or elastic wave equation, inversion starts with approximate initial values of subsurface parameters, which are typically updated in iterative fashion. Here, we propose a two-stage unsupervised machine-learning (ML) methodology for efficient and accurate seismic impedance inversion. The first stage utilizes the generalization capability of convolutional neural networks (CNN) to produce realistic estimates of the acoustic impedance (AI), whereas the second stage incorporates physics information to generate synthetic data from the subsurface AI distribution. We also add Bayesian layers to the first stage of the network to evaluate the model errors. The proposed probabilistic approach to deep learning allows one to estimate the uncertainty of the inverted parameters, which enhances the interpretability of the model. We apply the algorithm to a poststack data set generated using the CGG Hampson- Russell software. After conducting network training with a sufficient number of data points, the network is applied to the rest of the data to estimate the model parameters. The developed approach has a significant advantage over more conventional ML strategies because it produces statistically justified uncertainty maps and eliminates the need to use labeled data for training.