AI-ordlista
Den kompletta ordlistan över AI
Principal Component Analysis (PCA)
Linear dimensionality reduction technique that transforms correlated variables into uncorrelated components by maximizing explained variance along orthogonal axes.
t-SNE (t-Distributed Stochastic Neighbor Embedding)
Non-linear dimensionality reduction algorithm preserving local structures by minimizing the Kullback-Leibler divergence between probability distributions in the original and reduced space.
Isomap
Dimensionality reduction algorithm preserving geodesic distances by constructing a neighborhood graph and using multidimensional scaling.
MDS (Multidimensional Scaling)
Visualization technique preserving pairwise distances between points by finding a low-dimensional configuration that minimizes distance preservation stress.
Explained Variance
Proportion of total data variance captured by each principal component, serving as a criterion for selecting the optimal number of dimensions.
Autoencoders
Unsupervised neural networks learning compressed representations by forcing the output to reconstruct the input through a reduced-dimensional latent space.
Factor Analysis
Statistical method modeling observed variables as linear combinations of unobserved latent factors, separating common variance and unique variance.
t-SNE Perplexity
Hyperparameter controlling the effective number of neighbors considered in the t-SNE algorithm, influencing the balance between preservation of local and global structures.
ICA (Independent Component Analysis)
Blind source separation technique that seeks to decompose multivariate signals into statistically independent components by maximizing non-Gaussianity.
Variational Autoencoders
Probabilistic extension of autoencoders that learns a distribution in the latent space, enabling the generation of new data and better regularization.
PHATE (Potential of Heat-diffusion for Affinity-based Trajectory Embedding)
Algorithm that preserves trajectories and branches in data by combining heat diffusion and dimensionality reduction to visualize continuous processes.
NMF (Non-negative Matrix Factorization)
Matrix decomposition constrained to non-negative values, producing interpretable bases and additive representations of data.