Fig. 4

RMSprop optimizer provides the best reconstruction of the CVAE latent space for Fs-peptide simulations. We used four popular optimizers, including ADAM, ADAMax, ADAgrad and RMSProp for understanding how well we can reconstruct the latent space representation of the Fs-peptide simulations. For each optimizer, we show the reconstruction error, defined as the loss, \(\mathcal {L}\), for both the training/testing data over the course of 150 epochs. Notably, RMSProp provides the best reconstruction (lower is better) as indicated in the plot