Search papers, labs, and topics across Lattice.
Image Processing Laboratory (IPL), Universitat de València, Paterna, València 46980, Spain
2
0
5
Nonlinear $β$-VAEs reveal a surprising tradeoff where deeper models concentrate utility but compromise the fidelity of less significant dimensions.
Posterior collapse isn't a bug in VAEs, it's a feature that automatically prunes latent modes by importance, revealing a utility spectrum akin to PCA.
