AI-ordlista
Den kompletta ordlistan över AI
t-SNE (t-Distributed Stochastic Neighbor Embedding)
Non-linear dimensionality reduction algorithm particularly effective for visualizing high-dimensional data by focusing on preserving similar local structures.
Number of neighbors (n_neighbors)
Fundamental parameter of UMAP determining the size of the local neighborhood used to build the manifold representation, directly affecting structure preservation.
Repulsive force
Mechanism in visualization algorithms like t-SNE and UMAP preventing point clustering and preserving the local structure of data.
Gradient descent optimization
Iterative process used to minimize the cost function in t-SNE and UMAP, progressively adjusting point positions in the reduced space.
Local vs global structure
Fundamental trade-off in dimensionality reduction algorithms between preserving close neighborhood relationships and the global arrangement of clusters.
Topological manifold
Mathematical concept underlying UMAP assuming that high-dimensional data resides on a lower-dimensional surface embedded in the original space.
k-nearest neighbors graph (k-NN)
Intermediate data structure used by UMAP to model neighborhood relationships before projection into the reduced-dimensional space.
Min_distance
UMAP parameter controlling the minimum allowed distance between points in the reduced space, influencing the compactness of resulting clusters.
Embedding quality
Measure of the fidelity with which the low-dimensional representation preserves the structural relationships of the original data.
Stochastic embedding
Probabilistic nature of t-SNE where final positions can vary between runs, unlike deterministic dimensionality reduction approaches.
Cross-entropy optimization
Alternative to Kullback-Leibler divergence used in some t-SNE implementations for more stable optimization convergence.