PRML閱讀記(2)

繼續PRML

第二章：

Binary Variables:

• Bernouli Distribution, binomial distribution
• conjugate prior --> beta distribution

Multinomial Variables:

• multinomial distribution
• conjugate prior --> Dirichlet distribution

The Gaussian Distribution:

• univariate, multivariate, shape, limit
• conditional Gaussian, Marginal Gaussian, Bayes' theorem
• Maximum likelihood, sequential estimation
• conjugate prior --> unknown mean, unknown variance, and both
• Mixtures of Gaussians

The exponential Family:

• Bernoulli distribution --> logistic sigmoid function
• Multinomilal distribution --> softmax function
• conjugate priors

Nonparametric Methods:

• histogram approach: p(i) = n(i) / (N * width of bin)
• p(x) = K/NV:   fix V and find K --> Kernel Estimator . fix K and find V --> K-nearest-nerghbour estimator
• Kernel Estimator: estimate new data x according to old datas in V
• K-nearest-nerghbour estimator: estimate new data x according to neighbors within K

import views:

1. posterior = prior * ML

2. conjugate prior

3. sequential model to deal with large dataset(update data with disgarding the old data)

4. Gaussian Distribution and its variation

5. nonparametic method

6. hyperparameter: to model the distribution of parameter