This chapter presents the theoretical and methodological foundations underpinning the application of autoencoderAutoencoder-based artificial intelligence to the authentication of porcelain decoration. It positions the study within the broader history of computational art analysis, reviewing the evolution from early feature-based statistical methods to contemporary deep learningDeep learning architectures. Particular attention is given to one-class and anomaly detection approaches, which are especially suited to domains characterised by limited authenticated data, such as historical porcelain decoration. The chapter justifies the selection of convolutional autoencodersAutoencoder and details the hybrid methodological strategy adopted, combining carefully engineered handcrafted featuresHandcrafted features with neural network-based representation learning. It also examines challenges specific to porcelain authentication, including data scarcity, photographic variability, material degradation, collaborative production practices, and stylistic evolution over an artist’s career. By articulating both the scientific principles of autoencodersAutoencoder and the art-historical constraints of the problem domain, the chapter establishes a rigorous, interpretable, and context-sensitive framework for AI-assisted attribution.