[proba]处理混合模型的复杂度.

[shouldsee]

• Ian Goodfellow: https://arxiv.org/abs/1701.00160
  Figure 9: : Deep generative models that can learn via the principle of maximim likelihood
  differ with respect to how they represent or approximate the likelihood. O

• Hinton on MOE: https://www.cs.toronto.edu/~hinton/csc321/notes/lec15.pdf

• Topological data analysis http://topology.nipissingu.ca/workshop2018/slides/Goldfarb-Data-Beamer.pdf

混合模型的几大度量:

1. 拟合度
2. 参数的对称性/ identifiability
3. 参数的可复现性．