- From course Probabilistic Models and Inference Algorithms for Machine Learning, Prof. Dahua Lin
- All contents here are from the course and self understandings.
基本步骤
- 理解你要求解的某个问题
- what kind of entities/factors are involved?
- How do they interact with each other?
- Any constraints to take into account?
- 建立模型
- Introduce variables
- Specify relations among them: make assumptions and modeling choices
- Formalize the graphical model
- Derive the inference & estimation algorithms
Gaussian Mixture Model (GMM)
1. Motivation & Assumptions
- Observation: clusters (如图中所示)
- Assumptions:
- latent components
- independent generation of components (一个component指一个如图所示的cluster)
- independent generation of points in each component.
2. Formulate Gaussian Mixture Model
- Variables:
- sample points: xi
- component indicators: zi
- Generative procedure: for each i
- choose a component: zi∈π
- generate a point: xi∈N(μzi,Σzi) (每个sample point 服从正态分布,mean和方差为 μzi 和 Σzi )
- Model parameters:
- component parameters: {(μk,Σk)}k=1:K
- choice prior: π=(π2,⋯,πK)
- Joint distribution:
p(X,Z|Θ)=∏i=1NpC(zi|π)pN(xi|μzi,Σzi) - 该模型的图形表达如下:
- 但是以上的模型不能泛化:当有多个groups 的数据时候,每一个 group Gm 都有一个 prior: πm ,这样每一个group一个先验,对于新来的数据,并没有一个generalized的 πnew . 因此需要一个 Group-wise 的GMM
3. Extend Gaussian Mixture Model
- 一般来说,要泛华某个先验(prior),我们可以采用一些常见的分布,如下,这里我们采用Dirichlet
- Group-wise GMM: Generalizable to New Groups
- Introduce a Dirichlet Prior over πm to allow the generation of new groups.
- Formulation:
-
- For each group Gm:πm∼Dir(α) .
-
- Generate the i-th point in
Gm
:
zi∼πmxi∼N(μk,Σk)
- Generate the i-th point in
Gm
:
-
- 注意
πm
现在是个 隐变量 (latent variable)
- 注意
πm
现在是个 隐变量 (latent variable)
- Temporal Structures:
- 在实际世界中,时域上的变化是很常见的,那对于模型来说,也需要对应的dynamics作用到 不同的变量上
- Three ways to incorporate dynamics:
Dynamics on xi
Dynamics on zi
Dynamics on π