/nalu_implementation

This repository is created to show the Neural Arithmetic Logic Unit implementation in python using Tensorflow. The code in this repo complements my article on Medium on NALU

Primary LanguagePython

Simple implemtation of Neural Arithmetic Logic Units (NALU) in python using Tensorflow

The research engineers at DeepMind including well known AI researcher, Andrew Trask have published an impressive paper on a neural network model that can learn simple to complex numerical functions with great extrapolation (generalisation) ability.

alt text

This repository is created to show the Neural Arithmetic Logic Unit (NALU) implementation in python using Tensorflow. The code in this repo complements my article on Medium on NALUs. If you are new to NALUs, I strongly recommend my post for simple and intuitive explanation.

I have added three python files here:

  • nac_simple_implementation.py -> It implements the simple Neural Accumulator in python that is able to learn the addition and subtraction functions.
  • nac_complex_implementation.py -> It implements the complex Neural Accumulator architecture in python that is able to learn the complex arithmetic functions such as multiplication, division and power function.
  • nalu_implementation.py -> It implements the complete NALU using the same NAC implementation given in previous two files.

Note: I have also added the code for training and evaluation on test data in each file for completeness

I use the random uniform intiializer to initialize unconstrained parametrs W_hat and M_hat within the range [-2,2] . However one may use any recommended weight initializers as he want.

W_hat = tf.get_variable(name = "W_hat", initializer=tf.initializers.random_uniform(minval=-2, maxval=2),shape=[in_features, out_units],  trainable=True)

M_hat = tf.get_variable(name = "M_hat", initializer=tf.initializers.random_uniform(minval=-2, maxval=2), shape=[in_features, out_units], trainable=True)

For training and test data I generate the series of integers for both input x1 and x2 using numpy's arange API.

# Generate a series of input number X1 and X2 for training
x1 = np.arange(0,10000,5, dtype=np.float32)
x2 = np.arange(5,10005,5, dtype=np.float32)

# Prepare the input vector for training
x_train = np.column_stack((x1,x2))

# Generate a series of input number X1 and X2 for testing
x1 = np.arange(1000,2000,8, dtype=np.float32)
x2 = np.arange(1000,1500,4, dtype= np.float32)

# Prepare the input vector for testing
x_test = np.column_stack((x1,x2)

For the simple NAC I evaluate the addition operation on the generated training data y = x1 + x2. Whereas, for the complex NAC I use simple multiplication function y = x1 * x2 to evaluate the netwrok. For NALU, I use the complex numeric function y = (x1/4) + (x2/2) + x3**2 to evaluate the network.

Note: Readers are welcome to tune my model on various arithmetic functions and let me know the results.