2
$\begingroup$

It seems like Spectral Clustering is just a term for dimension reduction via Laplacian Eigenmaps + a clustering algorithm on the output.

Is this the case, or am I missing some fundamental difference?

$\endgroup$
2
$\begingroup$

Though choices of Similarity measure, Laplacian, and Clustering algorithm etc may differ in specific implementations, the general algorithm for Spectral Clustering is indeed identical to that of Laplacian Eigenmaps + a Clustering method:

4.6 Relation between LE and Spectral Clustering

Although Laplacian Eigenmaps are oriented to dimensionality reduction algorithm and Spectral Clustering is focused on cluster data, they are very related.

Both methods are approaches provided by the eigenvectors of Laplacian Graphs. Specifically, if we want to compare them, we must focus our attention on Normalized Spectral Clustering based on Random Walks (algorithm 5). If we compare both algorithms (see comparative table 4.6.1), we will see that the only difference is that Spectral Clustering makes one step more to cluster the information in the new dimension.

enter image description here

As one can see, steps 1 - 8 are identical for both algorithms, the only difference being whether the embedding itself is returned (LE) or a clustering algorithm is applied and the clusters returned (SC).

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.