伯努利分布(Bernoulli distribution):单次(only once)随机实验,结果有两种可能:事件 0 0 0(失败)和事件 1 1 1(成功)。假设 X ∈ { 0 , 1 } X \in \{ 0, 1 \} X∈{0,1}为二值随机变量(binary random variable), x = 1 x = 1 x=1的概率为 θ \theta θ、 x = 0 x = 0 x=0的概率为 1 − θ 1 - \theta 1−θ,则 X X X服从伯努利分布,记为 X ∼ Ber ( θ ) X \sim \text{Ber} (\theta) X∼Ber(θ),其概率累积函数(probability mass function,pmf)定义为:
Ber ( x ; θ ) = θ x ( 1 − θ ) 1 − x \text{Ber} (x ; \theta) = \theta^{x} (1 - \theta)^{1 - x} Ber(x;θ)=θx(1−θ)1−x
即
Ber ( x ; θ ) = { θ , if x = 1 1 − θ , if x = 0 \text{Ber} (x ; \theta) = \begin{cases} \theta, & \text{ if } x = 1\\ 1 - \theta, & \text{ if } x = 0\\ \end{cases} Ber(x;θ)={θ,1−θ, if x=1 if x=0
多项伯努利分布(multinoulli distribution):单次(only once)随机实验,结果有 K K K种可能。假设 x = ( x 1 , ⋯ , x K ) \mathbf{x} = (x_{1}, \cdots, x_{K}) x=(x1,⋯,xK)为随机向量(random vector), x j ∈ { 0 , 1 } x_{j} \in \{ 0, 1 \} xj∈{0,1}为二值随机变量, x j = 1 x_{j} = 1 xj=1的概率为 θ j \theta_{j} θj、 x j = 0 x_{j} = 0 xj=0的概率为 1 − θ j 1 - \theta_{j} 1−θj,则当事件 j j j发生时, x j = 1 x_{j} = 1 xj=1、 x i = 0 x_{i} = 0 xi=0( i ≠ j i \not = j i=j)。多项伯努利分布的概率累积函数为:
Multinoulli ( x ; θ ) = ∏ j = 1 K θ j x j \text{Multinoulli} (x ; \theta) = \prod_{j = 1}^{K} \theta_{j}^{x_{j}} Multinoulli(x;θ)=j=1∏Kθjxj