Table 3. Computational procedures of the contrastive divergence (CD) algorithm.

Algorithm: CD
Input: training set of N batch $v_{n=1}^N$, number of hidden units s, number of visible units nm
Output:bi, cj, wij1 Initialize:bi = 0, cj = 0, wij was randomly initialized, k = 50, ɛ = 0.12 fort = 1, 2, …, kdo3for each batch vn, n=1 , 2, …, Ndo4forj = 1, 2, …, sdo     // from visible layer to hidden layer 5$P\left(h_j^{\ast }=1|v\right)=\sigma \left(w_{ij}{v}_i+c_j\right)$6end for7fori = 1, 2, …, nmdo // from hidden layer to visible layer 8$P\left(v_i^{\ast }=1|h\right)=\sigma \left(w_{ij}{h}_j+b_i\right)$9end for10$w_{\mathrm{ij}}=w_{ij}+\varepsilon \ast \left[p\left(h_i=1|v\right)v_j^T-p\left(h_i^{\ast }=1|v^{\ast }\right)v_j^{\ast T}\right]$11$b_j=b_j+\varepsilon \left(v_j-v_j^{\ast }\right)$12$c_i=c_i+\varepsilon \ast \left[p\left(h_i=1|v\right)-p\left(h_i^{\ast }=1|v^{\ast }\right)\right]$13end for14 end for