2D distribution data can be predicted by Gaussian Process regression
updated on Dec 12th, 2019
Approx_Surface.m is a draft program for any developing.
(Finished)
Approx_Surface_GPR.m is the very conventional Gaussian process regression program [Rasmussen 2003].
Approx_Surface_poly55.m uses the fitting toolbox of MATLAB for method comparison.
Approx_Surface_GPR_datasetReduction.m works with datasetReduction.m, which naively reduces the data by step.
Approx_Surface_SVGP.m is going to implement the SVGP method [Titsias 2009].
datasetReduction.m is for reducing the dataset size in a naive way.
Doc. “beautiful Gp demo code” is coded for my presentation to show how GP works and how sparse approximation works.
Doc. “Local modeling project” is a project to train a GP-based error distribution prediction model for soil loading of excavator.
(On-going works)
Doc. “PCA study” is a project to reduce dimension of the representation of soil shape data by applying Principal component analysis (PCA) method.
Doc. “unit-symmetric Gpdf” is a project to use the skewed 2d Gaussian pdf to replace the nominal Gaussian pdf in our estimation.
Rasmussen C E. Gaussian processes in machine learning[C]//Summer School on Machine Learning. Springer, Berlin, Heidelberg, 2003: 63-71.
Burt D R, Rasmussen C E, Van Der Wilk M. Rates of Convergence for Sparse Variational Gaussian Process Regression[J]. arXiv preprint arXiv:1903.03571, 2019.
Titsias M. Variational learning of inducing variables in sparse Gaussian processes[C]//Artificial Intelligence and Statistics. 2009: 567-574.
