I am working on a project which aims to modify a neural network called the Self-Organizing Map SOM - which is essentially a clustering/dimensionality reduction algorithm that preserves the topology of given data. My goal is to propose several alternatives/variants of SOM which add a notion of relevance/ranking (in the context of relevance scores) so that high scoring images appear on top while preserving the local similarity still.
So in short, my dataset has 20,000 feature vectors and query scores of each of those vectors for 327 queries. And I wanna change the neural network so that in considers the relevence/ranking as well as the local similarity.
Do you have any ideas how I could go about this by chance?
Thanks a lot!