Shallow efficient learning with backpropagation in the space of functions

The estimation of functions of functions (nested functions) is one of the central reasons for the popularity that machine learning affords today. Here we introduce Representational Gradient Boosting (RGB), a non-parametric algorithm that estimates nested functions with multi-layer architectures obtained using backpropagation in the function space. RGB implications on meta-learning and time series analysis are discussed.

Speaker: Gilmer Valdes, PhD, Assistant Professor, Department of Radiation Oncology & Biostatistics, UCSF: Departments of Statistics & Oncology, Stanford University