Machine Learning (ML)

Machine learning differs from classic programming in one central point: instead of coding fixed rules, the system learns from examples. A spam filter is not programmed with rules like contains the word Viagra but with thousands of examples of spam and not-spam, from which it derives the relevant patterns itself.

ML splits roughly into three categories: Supervised Learning (with labeled training data), Unsupervised Learning (pattern recognition in unlabeled data), and Reinforcement Learning (learning through reward in interactive environments). Each category has its use cases, methods, and requirements for data volume and quality.

Pimcore integrates ML functions directly into production workflows. Product classification (supervised), anomaly detection in data quality (unsupervised), AI-driven enrichment, and agentic workflows (hybrid approaches) are part of native platform functions. Through the Agent SDK, custom ML models can be embedded in Pimcore workflows with the Data Spine as the data foundation.