/Neural_Networks

Whitebox and Blackbox implementation of Artificial Neural Networks

Primary LanguageJupyter Notebook

Artificial Neural Network [ANN]

An attempt to build a whitebox and blackbox implementation of Artificial Neural Networks to undersatnd how ANN learns.

Data

$x_{1}$ $x_{2}$ y
1 0 1
1 1 1
0 1 0
0 0 0

relation : y = $x_{1}$

2 Input variables $x_{1}$ , $x_{2}$
1 Output variable y

Neural Net Design

Hidden layers : 1 (2 Neurons)

Activation Function Used

  • Hidden Layer : Sigmoid
  • Output Layer : Sigmoid

Results

As expected the blackbox results were more accurate than whitebox. Whitebox results were biased towards output 1 even though it seemed to be learing and updating its weights through back propagation.