E-Graph Library Learning Models Jazz Harmony Patterns

A new computational model uses library learning on e-graphs to discover generative explanations of jazz harmonic patterns. The model, presented in a paper on arXiv (2605.04622), searches a program space of primitive harmonic relations to find concise programs for each piece in a corpus, then jointly learns a library of patterns and refactored programs. Deductive parsing integrated with e-graphs efficiently navigates the joint space. The study evaluates how well the model captures aspects of human musical pattern learning by testing the intuitiveness of programs and libraries.