Improving quality of generated images through an introspective manner. Combines the high quality generations of GANs while maintaining a latent representation of the images. Please see below for my analysis!
Introduces disentanglement into the VAE structure, throught a very simple tuning of a parameter, β. β controls the effect of the regularization term, which can constrain the latent space. Disentanglement aims to increase robustness and interpretability in these neural network models.
Compression of images into a vector representation. VAEs allow clustering of similar images in space. Can also randomly generate images. Maps the input space of images onto a very low dimensional space. For an analysis please see below!