PAC-Bayesian Reconstruction Guarantees for Time Series Variational Autoencoders

The researchers proposed a latent variable model for time series, based on the PAC-Bayesian framework. The study extended the PAC-Bayesian guarantees to the Markov latent structure, utilized sequential generation processes to capture temporal dependencies, and demonstrated that these guarantees do not increase with the length of the trajectory. This framework relies on assumptions commonly found in the literature, and provides non-restrictive examples to verify these assumptions.