AI-ordlista
Den kompletta ordlistan över AI
Upper Confidence Bound (UCB)
Acquisition function that adds an exploration term, proportional to the model's uncertainty, to the predicted mean, allowing explicit control of the exploration-exploitation trade-off via a parameter.
Probability of Improvement (PI)
Acquisition function that calculates the probability that a candidate point will improve the current best result, tending to favor exploitation of areas already known to be promising.
Kernel (or Covariance Function)
Function that defines the similarity between two points in the search space and determines the smoothing properties of the Gaussian process, crucial for the quality of the surrogate model.
Multi-Objective Optimization
Extension of Bayesian optimization aimed at simultaneously optimizing multiple conflicting objectives, producing a Pareto front of non-dominated solutions.
Parallel Optimization
Variant of Bayesian optimization that proposes multiple points to evaluate simultaneously, accelerating the process by leveraging distributed computing resources.
Kernel Hyperparameter
Parameters of the Gaussian process kernel (such as length scale or variance) that define the correlation structure of the surrogate model and are often optimized internally.
Latin Hypercube Sampling Initialization
Strategy for generating the first evaluation points that ensures homogeneous coverage of the search space, reducing the initial bias of the surrogate model.
Trust Region Based Optimization
Approach that restricts the search to a trust region around the current best solution, dynamically adjusting it to accelerate local convergence.
Noisy Objective Function
Evaluation function whose results are affected by random noise, requiring adaptations of the Gaussian process and acquisition functions to handle uncertainty.
Automated Hyperparameter Tuning
Process that uses algorithms such as Bayesian optimization to automatically find the combination of hyperparameters that optimizes the performance of a machine learning model.