【个人向】计算建模 补充材料

计算建模是HIT大三秋季计院计算机科学方向的专业核心课，内容cover了随机过程、优化、信号处理等topic
这里是笔者对课余参考的一些补充材料的collection，主要是一些概念
个人向的意思是，本文不对内容的严谨性与正确性负责

Bayesian Inference

MLE & MAP : 略

Stochastic Processes

A random process(stochastic process) is a collection of random variables usually indexed by time.
A random process is a random function of time.

Stationary

Similar definitions (Strict-sense stationary / WSS) apply for discrete-time random process.
WSS is more commonly observed and used.

Gaussian Process

Gaussian processes can be seen as an infinite-dimensional generalization of multivariate normal distributions.

HMM

Definition

Probabilities Calculation

• given model $M=(A,B,\pi )$ (where $A$ is the transition matrix, $B$ is the emission matrix, $\pi$ is the initial value vector) and observation sequence $Y$

forward & backward DP:

• ${\alpha }_t(i)=P(Y_1,Y_2,\dots ,Y_t,X_t=q_i|M)$
• ${\beta }_t(i)=P(Y_t,Y_{t+1},\dots ,Y_T|X_t=q_i,M)$

Model Learning

• EM algorithm

Decoding

• Viterbi algorithm (DP)

Random Simulation

假设已有一个均匀分布的随机数，如何由它得到任意连续分布的随机数？
设我们想要的服从某种特定连续分布的随机数为$X$，它的CDF为$P_X(x)=P[X\le x]$
令$U=P_X(X)$，可以证明$U$服从$[0,1]$上的均匀分布
考虑$P[U\le y]=P[P_X(X)\le y]$
设$P[P_X(x_0)=y]$，由$P_X$的单调性知，$P[P_X(X)\le y]=P[X\le x_0]=P_X(x_0)=y$
即$P[U\le y]=y$，即$U\sim U[0,1]$
那么只需要把$U=P_X(X)$取反函数即可得到$X=P_X^{-1}(U)$

Reference

Introduction to Probabilities, Statistics and Random Processes
Wikipedia