AI-ordlista
Den kompletta ordlistan över AI
Graph Neural Network
Deep learning architecture designed to process structured graph data, enabling learning of representations from relationships between nodes and edges.
Spatial GNN
GNN approach that performs convolution directly on the neighborhood space by aggregating features of adjacent nodes according to predefined schemes.
Spectral GNN
Family of GNNs based on spectral graph theory, using graph Fourier transform to define convolution operations.
Graph Fourier Transform
Generalization of the classical Fourier transform to signals defined on graphs, using the eigenvectors of the Laplacian matrix.
Laplacian Matrix
Square matrix representing the structure of a graph, essential in spectral analysis and defined as L = D - A where D is the degree matrix.
Spectral Filter
Filter applied in the spectral domain to modify the frequencies of a graph signal, analogous to filters in classical signal processing.
Aggregation Function
Mathematical operation in spatial GNNs that combines features of neighboring nodes, such as mean, max, or sum pooling.
Chebyshev Polynomials
Orthogonal polynomials used to efficiently approximate spectral filters in GNNs, reducing computational complexity.
Eigendecomposition
Fundamental matrix decomposition for spectral GNNs, computing eigenvalues and eigenvectors of the Laplacian matrix.
Spectral Clustering
Partitioning method using eigenvalues of the Laplacian matrix to identify communities in graphs.
Graph Isomorphism Network
Powerful spatial GNN architecture with strong theoretical expressivity properties, capable of distinguishing most non-isomorphic graphs.
Graph Wavelet Transform
Alternative to graph Fourier transform offering simultaneous spatial-spectral localization for multi-scale analysis.
Neighborhood Sampling
Neighbor sampling strategy in spatial GNNs to handle large graphs and reduce computational complexity.
Spectral-Temporal GNN
GNN extension combining spectral analysis and temporal modeling to process evolving dynamic graphs.