Volume 9 - Issue 12
Sparse gradients learning model
Abstract
Recent studies have shown that gradients of the target function are very useful measure to discover the inner relationships between the variables and the influence of the features on the output data in high-dimensional data analysis tasks. Inspired by the state-of-the-art Gaussian Process Gradient Learning (GPGL) model and Relevance Vector Machine (RVM) we propose a novel gradients learning model in bayesian framework, named Sparse Gradients Learning (SGL) model, which can not only provide the error bars for the estimated gradients but also improve the accuracy of the gradients estimation significantly because sparse model has high generalization capability. Simulated experiments verify the proposed model.
Paper Details
PaperID: 84880350169
Author's Name: Hou, Z., Huang, S., Zhang, Z.
Volume: Volume 9
Issues: Issue 12
Keywords: Gradients learning, Sparse model, Variable selection
Year: 2013
Month: June
Pages: 4655-4662