Generalized Matheron Variational Implicit Processes

Key points

• Implicit-process priors specify distributions over functions through sample-forward mechanisms such as Bayesian neural networks and stochastic simulators, but their function-space densities are typically unavailable.1
• For Gaussian-process priors, GMVIP recovers the standard inducing-variable variational GP construction; for general implicit priors, its empirical covariance construction preserves the prior mean and covariance in the population limit.1
• GMVIP constructs posterior samples by drawing a function from the prior and applying a correction anchored at a set of inducing inputs.1
• Experiments on regression, classification, and forecasting with simulator-defined and retrieval-conditioned empirical trajectory priors show that GMVIP is broadly competitive with existing methods.1

Sources (1)

• [1]Generalized Matheron Variational Implicit Processes
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 08:11 AM

  We introduce Generalized Matheron Variational Implicit Processes (GMVIP), a pathwise variational family for posterior inference with such priors.

  Implicit-process priors specify distributions over functions through sample-forward mechanisms such as Bayesian neural networks and stochastic simulators, but their function-space densities are typically unavailable.

Extractive summary: sentences quoted from the sources.