GP-pro-c
A Gaussian process surrogate model based on a product of experts, which adds an uncertainty calibration step to correct the overconfidence of the standard product-of-experts formulation. Designed for large-scale Bayesian optimization, it can be combined with trust-region mechanisms to provide more robust acquisition-function decisions.
Timeline (last 90 days)
The paper proposes an algorithm named BO-pro-c-tr that uses a product-of-experts Gaussian process with uncertainty calibration (GP-pro-c) to重构 the paradigm of large-scale Bayesian optimization.
The "one-to-many association strategy" proposed in the paper allows a single GP-pro-c surrogate model to support multiple trust regions, thereby decoupling the scheduling layer from the modeling layer.
All Facts (2)
The paper proposes an algorithm named BO-pro-c-tr that uses a product-of-experts Gaussian process with uncertainty calibration (GP-pro-c) to重构 the paradigm of large-scale Bayesian optimization.
50%UnverifiedThe "one-to-many association strategy" proposed in the paper allows a single GP-pro-c surrogate model to support multiple trust regions, thereby decoupling the scheduling layer from the modeling layer.
50%