Algorithm. Contrastive Divergence (CD算法)

Input: RBM（ $V_1,\cdots ,V_m,H_1,\cdots ,H_n$ )，training batch S

Output： $\Delta w_{ij},\Delta b_j,\Delta c_i$ , for $i=1,\cdots ,n$ ; $j=1,\cdots ,m$

init $\Delta w_{ij}=\Delta b_j=\Delta c_i=0$ for $i=1,\cdots ,n$ ; $j=1,\cdots ,m$

for all the $v\in S$ do

$v^{\left(0\right)}\leftarrow v$ do

for $t=0,\cdots ,k-1$ do

for $i=1,\cdots ,n$ do sample

for $j=1,\cdots ,m$ do sample $v_j^{\left(t+1\right)}~p\left(v_j|h^{\left(t\right)}\right)$

for $i=1,\cdots ,n$ , $j=1,\cdots ,m$ do

$\Delta w_{ij}\leftarrow \Delta w_{ij}+p\left(H_i=1|v^{\left(0\right)}\right)\cdot v_j^{\left(0\right)}-p\left(H_i=1|v^{\left(k\right)}\right)\cdot v_j^{\left(k\right)}$

$\Delta b_j\leftarrow \Delta b_j+v_j^{\left(0\right)}-v_j^{\left(k\right)}$

$\Delta c_i\leftarrow \Delta c_i+p\left(H_i=1|v^{\left(0\right)}\right)-p\left(H_i=1|v^{\left(k\right)}\right)$