[ALGO] Mixture of Gaussian Process for unsupervised clustering

[shouldsee]

Mixture of Gaussian Distribution is well developed with EM/Variational/MCMC algorithms for inferring parameter, with high-performance open-sourced implementations available. In contrast, Mixutre of Gaussian Processes seems to be much less popular and a search over Github does not yield any well-maintained project. Whether this is due to theoretical or computational reasons is up to discussion.

Resources:

Mixture of Gaussian processes

• http://proceedings.mlr.press/v28/ross13a.pdf
• Nice slide on GP: http://mlss2011.comp.nus.edu.sg/uploads/Site/lect1gp.pdf
  • Complexity of O(M^2N), compare to Li and Stephen’s model https://arxiv.org/pdf/1006.1514.pdf
• Gibbs-sampling MCMC from Gatsby Institute with Cam CS (http://mlg.eng.cam.ac.uk/zoubin/papers/iMGPE.pdf)