Spectral Clustering

1. Create neighbourhood graph
2. Compute laplacian
3. Compute eigenvectors of laplacian, the ones with the smallest eigenvalues give new features
4. Run k-means clustering on new features