AGML 2012 - Algebraic Geometry in Machine Learning 2012

For many classes of machine learning models, theoretical understanding has lagged behind experimental success. In many cases, representational power and performance characteristics are poorly understood, and even proponents are unsure why they work.
Understanding the algebraic, polyhedral, and tropical geometry of graphical models and other popular models has provided a new set of tools enabling researchers to settle several open questions about their capabilities, and progress on this front is expected to continue.

Topics for the special session may include the algebraic geometry and representation theory of machine learning models, the polyhedral and tropical geometry of the space of functions they can compute, geometric characterizations of architecture choice and asymptotic performance, and related topics.