Fig. 3: Latent variables and independent components.
From: Inverse imaging with elastic waves driven by unsupervised machine learning
a The relationships between each pair of meaningful latent variables after VAE, colored by the magnitude of the height of the second block \({h}_{2}\), showing high correlation to latent variable \({z}_{2}\). b The relations between different {\({z}_{i}^{{\prime}}\)} (independent components after ICA) and {\({E}_{0}\), \({h}_{1}\), \({h}_{2}\), \({h}_{3}\)}, with the digits on the top-right corner representing Pearson correlation coefficients measuring the strength of linear dependence, indicating that one physical variable is mapped to only one independent component. c The relations between the recovered physical parameters \(\{{E}_{0}^{{\prime} },{h}_{1}^{{\prime} },{h}_{2}^{{\prime} },{h}_{3}^{{\prime} }\}\) from our ML model and their real values \(\{{E}_{0},{h}_{1},{h}_{2},{h}_{3}\}\) (ground truth), for the testing set of simulation data, with the digits on the bottom-right corner indicating the mean absolute errors.
