Deep Clustering and Interpolation via the Federated Self-Organizing Map
Keywords: self, organizing, map, federated, cybersecurity, security, ember, interpolation, clustering
TL;DR: The paper describes a federated self-organizing map for deep clustering and interpolation of tabular data, notably cybersecurity data.
Abstract: We introduce FedSOM, a clustering and interpolation module based on the Self-organizing Map (SOM), which can be appended to any encoder and which can be trained in a federated way either in tandem with the encoder or post training on the resulting representations. The result is a discrete moduli space of representations that provides for cluster or sample-level interpolation, hierarchical clustering, and can be leveraged as a function to cluster new vectors at test time. This moduli space can either be created from data alone or by glueing pre-existing clusters along regions of commonality, although we do not explore the latter in this work. Interpolation is accomplished by considering the $n$-dimensional tensor underlying the SOM as a weighted undirected graph, where the weights are computed as a function of the dispersion of the two clusters corresponding to the nodes bounding the given edge. Any two clusters or samples may then be interpolated by computing the lowest-cost path between their associated graph nodes via Dijkstra's algorithm. The method is validated on MNIST-like and parsed-binary malware datasets.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 1684
Loading