Are Kohonen Self-Organizing Maps or similar still a thing?

Reading about autoencoding in this article linked from this forum post being a form of unsupervised learning, triggered a memory from 1990s where I had a brief undergrad experience writing code for a Kohonen Self-Organizing Map:

I’m wondering how Kohonen maps specifially and SOMs in general fit into the current machine learning ecosystem?
i.e. is it still relevant? or has evolved into something else? or does DL provide an equivalent or superset behaviour?

1 Like

I haven’t seen them mentioned for decades!


Same here, the last time I worked with Kohonen maps was during MATLAB classes at the university. Also, I remember that some time ago, it was mentioned in our corporate ML newsletter. But don’t know if it used somewhere in the products.

Can’t say for sure, but does it resemble modern diffusion models? Like, starting from normal distribution and fitting it to the data? Or mapping the data into latent representation?

1 Like