| { |
| "field_id": "F025", |
| "field": "bayesian optimization", |
| "directions": [ |
| { |
| "direction_id": "G0", |
| "name": "Information-oriented experimental design", |
| "scope": "Select experiments that help learn or distinguish scientific models, and study the formulation, computation, and optimization of information-oriented design criteria, including when models may be misspecified." |
| }, |
| { |
| "direction_id": "G1", |
| "name": "Optimization guided by prior tasks and external knowledge", |
| "scope": "Use information available beyond evaluations of the current objective function to form search priors, surrogate models, policies, or query guidance for more efficient optimization of a new task. Such information may come from related optimization tasks or from experts, users, or decision-makers." |
| }, |
| { |
| "direction_id": "G2", |
| "name": "Surrogate modeling for complex optimization problems", |
| "scope": "Build surrogate predictions for Bayesian optimization that can support decisions by representing the structure of search domains and objective relationships while making predictive distributions, uncertainty inference, and computational capacity reliable and usable." |
| }, |
| { |
| "direction_id": "G3", |
| "name": "Localization and dimension reduction of search", |
| "scope": "Concentrate expensive optimization evaluations in promising local regions or along influential parameter directions, adapting that focus as the search proceeds." |
| }, |
| { |
| "direction_id": "G4", |
| "name": "Optimization with heterogeneous evaluation evidence", |
| "scope": "Use observable evaluations that are related to the target but differ in cost or content, assessing their value for the optimization target to improve search efficiency." |
| }, |
| { |
| "direction_id": "G5", |
| "name": "Acquisition and evaluation scheduling", |
| "scope": "Decide where and when to conduct the next expensive evaluation or batch of evaluations, addressing exploration and exploitation, redundancy among parallel evaluations, and feedback that has not yet arrived." |
| }, |
| { |
| "direction_id": "G6", |
| "name": "Reliable optimization under uncertain operating conditions", |
| "scope": "Make optimization decisions meet reliability goals when feasibility, safety boundaries, implemented inputs, or external contexts are uncertain. This includes both how evaluations should be conducted during the search and how the final decision should perform." |
| }, |
| { |
| "direction_id": "G7", |
| "name": "Multi-objective trade-off discovery", |
| "scope": "Discover or characterize Pareto trade-offs and explicitly specified multi-objective utilities across competing objectives, rather than optimizing only a single fixed scalar outcome." |
| }, |
| { |
| "direction_id": "G8", |
| "name": "Comparative assessment of Bayesian optimizers", |
| "scope": "Establish comparable evidence of how Bayesian optimization methods perform across different problems through benchmarks, controlled comparisons, or analysis of algorithm components." |
| } |
| ] |
| } |
|
|
