Unsupervised Machine Learning

Unsupervised learning addresses a practical bottleneck in ML: labeled training data is expensive and slow to produce, unlabeled data is plentiful. Instead of having humans manually categorize every data point, an unsupervised model learns the inner structure of the data and uses it to deliver insight.

The main applications are clustering (grouping data points into natural groups without the groups being known upfront), anomaly detection (identifying data points that deviate significantly from the norm), dimensionality reduction (compressing high-dimensional data into a few meaningful dimensions), and embedding computation (representing data as vectors in a semantic space).

Pimcore uses unsupervised methods in several function areas. Data quality checks identify conspicuous product data without anomaly patterns having to be labeled in advance, semantic product similarity finds comparable products through embedding spaces, automatic grouping proposes clusters for assortment analyses. Through the Agent SDK, custom unsupervised models can be embedded in workflows.