Decoding the shift-invariant data: applications for band-excitation scanning probe microscopy ^*

A shift-invariant variational autoencoder (shift-VAE) is developed as an unsupervised method for the analysis of spectral data in presence shifts along parameter axis, disentangling physically-relevant from other latent variables. Using synthetic sets, we show that shift-VAE variables closely match ground truth parameters. The shift VAE extended towards band-excitation piezoresponse force microscopy data, resonance frequency peak shape parameters a model-free manner. extensions this approach denoising and dimensionality reduction imaging spectroscopic are further demonstrated. This universal can also be to x-ray diffraction, photoluminescence, Raman spectra, sets.