Multifidelity bayesopt loop

As explained in Bayesian Optimization in Action

A multifidelity BayesOpt policy is a Bayesian optimization decision rule that selects query locations and fidelity levels by balancing information about the objective against evaluation cost.

A multifidelity BayesOpt loop finds an expensive objective’s optimum by choosing both where to query and which fidelity to use. It exploits inexpensive, imperfect approximations while reserving costly ground-truth evaluations for promising regions. section 9.0 section 9.1

The loop repeats three decisions:

• Train a Gaussian process on observations from both the high-fidelity ground-truth function and the low-fidelity approximation. section 9.1
• Score each data point–fidelity pair with an acquisition value that balances expected optimization progress against query cost. section 9.1 section 9.3
• Query the fidelity attached to the highest-scoring pair, add the observation to the training data, and repeat. section 9.1

Forrester-function search

A low-fidelity approximation of the Forrester function differs from the ground truth but follows roughly the same shape. The policy can sample the approximation broadly, then query the ground truth to refine the likely optimum. section 9.1

Figure 9.1 Model of a multifidelity decision-making problem, where the agent needs to balance querying the true objective function f(x) for accurate information and querying the inexpensive surrogate f̄(x)

Use the decision diagram to see the central trade-off between accurate, expensive ground-truth queries and inexpensive surrogate queries. section 9.1

How Bayesian Optimization in Action covers it

How deep it goes

Develops the loop through GP training, pairwise acquisition scoring, querying, and dataset updates, then defers cost modeling details.

Examples

Uses online reviews, neural-network training duration, physical simulations, and the Forrester function to illustrate fidelity and cost trade-offs.

What you can do after

Builds a cost-aware multifidelity BayesOpt policy that selects query locations and fidelity levels iteratively.

How it connects

Connects Gaussian processes, Bayesian optimization, acquisition functions, low-fidelity approximations, ground-truth objectives, and cost models.

liveBookmanning.com

Read Bayesian Optimization in Action on liveBook Get the book at manning.com

Related concepts

• Bayesian optimizationkind of · Bayesian optimization is a sequential method for optimizing an expensive or unknown objective by using a probabilistic model to guide evaluations.
• multifidelity optimizationaddresses · Multifidelity optimization combines evaluations of an objective at different levels of accuracy and cost to reduce optimization expense.
• multifidelity max-value entropy searchincludes · Multifidelity Max-value Entropy Search is a Bayesian optimization policy that selects both input locations and fidelity levels to reduce uncertainty about the objective's maximum while accounting for evaluation cost.

This explainer was generated by Manning's knowledge graph from the text of Bayesian Optimization in Action. The section links point to the parts of the book it draws on.