A Self-Organizing Map (SOM) is an unsupervised neural network that projects high-dimensional input data, such as CSI feature vectors, onto a lower-dimensional discrete grid of neurons through competitive learning, preserving the topological structure of the original data. In WiFi sensing, SOMs are valued for their ability to cluster and visualize complex, unlabeled CSI patterns — such as those associated with human activities or gestures — without requiring annotated training data, making them particularly useful in scenarios where ground-truth labels are scarce or costly to obtain. Key variants relevant to the field include growing SOMs and hierarchical SOMs, which dynamically adjust their topology to better accommodate the variability and non-stationarity inherent in real-world CSI measurements.
Dictionary term