Overcoming Mean-Field Approximations in Recurrent Gaussian Process Models
This work addresses a specific bottleneck in variational inference for dynamical systems, offering an incremental improvement over mean-field approximations.
The paper tackles the problem of miscalibrated posteriors and unnecessarily large noise terms in recurrent Gaussian process models by eliminating the factorisation between latent states and the transition function, resulting in better predictive performance and more calibrated estimates while maintaining computational efficiency.
We identify a new variational inference scheme for dynamical systems whose transition function is modelled by a Gaussian process. Inference in this setting has either employed computationally intensive MCMC methods, or relied on factorisations of the variational posterior. As we demonstrate in our experiments, the factorisation between latent system states and transition function can lead to a miscalibrated posterior and to learning unnecessarily large noise terms. We eliminate this factorisation by explicitly modelling the dependence between state trajectories and the Gaussian process posterior. Samples of the latent states can then be tractably generated by conditioning on this representation. The method we obtain (VCDT: variationally coupled dynamics and trajectories) gives better predictive performance and more calibrated estimates of the transition function, yet maintains the same time and space complexities as mean-field methods. Code is available at: github.com/ialong/GPt.