推导1

γt(i)\gamma_t(i)的推导

γt(i)\gamma_t(i)代表t时刻状态为i的概率。

用公式表达为:

γt(i)=P(it=qiO,λ)=P(it=qi,Oλ)P(Oλ)贝叶斯公式,P(AB)=P(A,B)P(B)=P(it=qi,Oλ)j=1NP(it=qi,Oλ)=αt(i)βt(i)j=1Nαt(i)βt(i)公式说明1\begin{aligned} \gamma_t(i) & = P(i_t = q_i | O, \lambda) \\ & = \frac{P(i_t = q_i , O| \lambda)}{P(O|\lambda)} & \text{贝叶斯公式,}P(A|B) = \frac{P(A, B)}{P(B)} \\ & = \frac{P(i_t = q_i , O| \lambda)}{\sum_{j=1}^NP(i_t = q_i , O| \lambda)} \\ & = \frac{\alpha_t(i)\beta_t(i)}{\sum_{j=1}^N\alpha_t(i)\beta_t(i)} & \text{公式说明1} \end{aligned}

公式说明

  1. 根据前向概率αt(i)\alpha_t(i)后向概率βt(i)\beta_t(i)的字义,得:

    αt(i)βt(i)=P(o1,o2,,ot,it=qiλ)P(ot+1,ot+2,,oTit=qi,λ)=P(o1,o2,,otit=qi,λ)P(ot+1,ot+2,,oTit=qi,λ)P(it=qiλ)=P(Oit=qi,λ)P(it=qiλ)=P(O,it=qiλ)\begin{aligned} \alpha_t(i)\beta_t(i) & = P(o_1,o_2,\cdots,o_t,i_t=q_i|\lambda) * P(o_{t+1},o_{t+2},\cdots,o_T |i_t=q_i, \lambda) \\ & = P(o_1,o_2,\cdots,o_t|i_t=q_i, \lambda) * P(o_{t+1},o_{t+2},\cdots,o_T |i_t=q_i, \lambda) * P(i_t=q_i|\lambda) \\ & = P(O|i_t=q_i, \lambda) * P(i_t=q_i|\lambda) \\ & = P(O, i_t=q_i| \lambda) \end{aligned}

    我的推导方法可能比较笨。

ξt(i,j)\xi_t(i,j)的推导

Last updated