-
-
Save omsrisagar/23c6bb6e55f56fbdd1a88985f972262c to your computer and use it in GitHub Desktop.
Neural Network Multiplication Approximator
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# Reference | |
# ========= | |
# Why does deep and cheap learning work so well?∗ | |
# Henry W. Lin, Max Tegmark, and David Rolnick | |
# Dept. of Physics, Harvard University, Cambridge, MA 02138 | |
# Dept. of Physics, Massachusetts Institute of Technology, Cambridge, MA 02139 and Dept. of Mathematics, Massachusetts Institute of Technology, Cambridge, MA 02139 | |
# Here the x input takes two numbers and produces the multiplication as output | |
import numpy as np | |
# multiplication approximator | |
lambda_val = 0.000001 | |
sigmoid = lambda x : 1./(1 + np.exp(-x)) | |
sigmoid_sec = lambda x: sigmoid(x)*(1 - sigmoid(x))*(1 - 2*sigmoid(x)) | |
bias1 = 2.3787 | |
# provide the input here, it works for two numbers as of now. | |
# It should not work for very large numbers. In that case, | |
# place lambda_val to a smaller value. But it does work really well | |
# till 1e4 ~ 1e5 range. | |
x = np.array([12, 45]).reshape(-1,1) | |
W1 = np.array([[lambda_val, lambda_val], [-lambda_val, -lambda_val], [lambda_val, -lambda_val], [-lambda_val, lambda_val]]) | |
b1 = np.array([bias1, bias1, bias1, bias1]).reshape(-1,1) | |
mu = 1/((lambda_val**2)*4*sigmoid_sec(bias1)) | |
W2 = np.array([mu, mu, -mu, -mu]) | |
z1 = np.dot(W1, x) + b1 | |
a1 = sigmoid(z1) | |
z2 = np.dot(W2, a1) | |
print(z2[0]) |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment