Researchers at Emory University propose a new mathematical framework that acts like a “periodic table” for AI, organizing machine-learning methods under one principle: compress data while preserving the most useful predictive information.

The model—called the Variational Multivariate Information Bottleneck—could guide algorithm design without heavy trial and error. If successful, it may improve multimodal AI while reducing the computing power and data needed to train future systems.

This episode includes AI-generated content.