Adaptive Learning Method of RBM

$$$$def expectation(l):

N=10000

return mean(l() for _ in range(N))

expectation(lambda: ("""boldmath v""", """boldmath h""")) ==

- sum(b_i v_i - sum(c_j h_j - sum(sum(v_i W_ij h_j))))

Adaptive Learning Method of RBM

$$$$def expectation(l):

N=10000

return mean(l() for _ in range(N))

p("""boldmath v""", """boldmath h""") == 1/Z * exp(- expectation(lambda:

("""boldmath v""", """boldmath h"""))), Z == sum(sum(exp(- expectation(lambda:

("""boldmath v""", """boldmath h""")))))

Neuron Generation and Annihilation Algorithm

$$$$(alpha_c @ dc_j) @ (alpha_W @ dW_ij) > theta_G

Adaptive Learning Method of RBM

$$$$def expectation(l):

N=10000

return mean(l() for _ in range(N))

expectation(lambda: ("""boldmath v""", """boldmath h""")) ==

- sum(b_i v_i - sum(c_j h_j - sum(sum(v_i W_ij h_j))))

Adaptive Learning Method of RBM

$$$$def expectation(l):

N=10000

return mean(l() for _ in range(N))

p("""boldmath v""", """boldmath h""") == 1/Z * exp(- expectation(lambda:

("""boldmath v""", """boldmath h"""))), Z == sum(sum(exp(- expectation(lambda:

("""boldmath v""", """boldmath h""")))))

Neuron Generation and Annihilation Algorithm

$$$$(alpha_c @ dc_j) @ (alpha_W @ dW_ij) > theta_G