Slowly Varying Regression Under Sparsity

Abstract

Building Sparse and Temporally or Spatially Linked Regression Models The framework of slowly varying regression under sparsity addresses many problems in machine learning where the underlying model is sparse and varies slowly and sparsely. This includes problems with temporally or spatially varying structures. For example, in the temporal case, the factors important in predicting the energy consumption in a building can vary depending on the hour of the day or the period of the year. In the spatial case, the factors that affect house prices can differ by neighborhood. Our paper rigorously formulates such models and proposes new theories, algorithms, and software for solving them. We thoroughly evaluate our framework on synthetic and real-world case studies and make our implementations available open-source to facilitate adoption by practitioners.