KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Multi-Objective Bayesian Optimization (MOBO)
Extension of Bayesian optimization using surrogate models to guide the search for a Pareto front with a minimum number of expensive evaluations.
Multi-Objective Gaussian Process
Surrogate model where each objective function is modeled by an individual Gaussian process, capturing uncertainty and correlation between objectives.
Multi-Objective Acquisition Function
Criterion exploiting the trade-off between exploration and exploitation to select the next point to evaluate, based on the predictions and uncertainties of the surrogate model.
Expected Hypervolume Improvement (EHVI)
Acquisition function that calculates the expected improvement in the hypervolume of the current Pareto front if a new point were evaluated.
Pareto Front Expected Improvement (PFEI)
Acquisition function that estimates the potential improvement of the Pareto front by evaluating a new point, based on the probability of non-dominance.
Multi-Objective Lower Confidence Bound (LCB) Criterion
Pessimistic acquisition function that optimizes a linear combination of the predicted mean and variance from the model for each objective.
Front Diversity Optimization
Strategy aimed at maintaining a good distribution of solutions on the Pareto front to avoid concentration in a single region of the objective space.
Scalarization
Technique transforming a multi-objective problem into a single-objective problem by weighting the different objectives, often used to define acquisition functions.
Multi-Objective Kriging
Synonymous with the use of Gaussian processes for multi-objective modeling, inherited from the geostatistics field where Kriging is an interpolation method.
Epsilon-Indicator
A quality metric that quantifies the worst performance of one set of solutions compared to another, measuring the factor by which one front must be degraded to dominate another.
Decomposition-Based MOBO
An approach that decomposes the multi-objective problem into several single-objective subproblems, each solved by standard Bayesian optimization.
Batch Bayesian Multi-Objective Optimization
A variant of MOBO where multiple points are selected simultaneously for evaluation, often in parallel, to accelerate the optimization process.