Automatic concept learning

Description

Can an AI learn the laws of music theory from sheet music in the same human-interpretable form as a music theory textbook? How little prior knowledge is needed to do so? Haizi Yu considers questions like these as he walks you through developing a general framework for automatic concept learning.

While concept learning is pervasive in humans, current AI systems are mostly good at either applying human-distilled rules (rule-based AI) or capturing patterns in a task-driven fashion (pattern recognition) but not at learning patterns in a human-interpretable way similar to human-induced concepts (Grenander’s pattern theory). Haizi targets both self-exploratory and self-explanatory AI that conceptualizes a domain, so people can use its distilled concepts to solve more specific tasks thereafter.

Built upon the core idea of abstraction and enabling an AI to make abstractions, he formalizes automatic concept learning as a generalization of Shannon’s information lattice, which itself encodes a hierarchy of abstractions and is algorithmically constructed from group-theoretic foundations. The core concept learning algorithm is realized by an iterative discovery cycle that has a student-teacher architecture and that operates on a generalized information lattice.

Haizi then offers an overview of a use case in the domain of music—an automatic music theorist MUS-ROVER that distills from sheet music compositional rules that resemble many music theory textbooks. The MUS-ROVER web application can further deliver personalized lessons on music composition.