Spectral clustering refers to all methods that divide a set of data into clusters using the eigenvectors of the adjacency matrix of a graph or other matrices derived from it. An adjacency matrix describes a complete graph with vertices in objects and edges between each pair of objects with a weight corresponding to the degree of similarity between these vertices. Spectral clustering is a transformation of the initial set of objects into a set of points in space whose coordinates are elements of eigenvectors. The formal name of such a task is the normalized cuts problem. The resulting set of points is then clustered using standard methods—for example, with the k-means algorithm. Changing the representation created by eigenvectors allows us to set the properties of the original set of clusters more clearly. Thus, spectral clustering can separate points that cannot be separated by applying k-means—for example, when the k-means method gets a convex set of points. The main disadvantage of spectral clustering is its cubic computational complexity and quadratic memory requirements.