S6

Welcome to the Notes explaining the major steps taken to emulate backpropagationon Excel for a NN with a single hidden layer

Steps for the Process

1. Layer Initialisation: The nodes of the output and hidden layer are filled using the initial parameter values of the 8 weights W1 till W8.
2. Loss Calculation: The error is calculated using the error loss functions (1/2)(t1 - Ao1) & (1/2)(t2 - Ao2).
3. BackPropagation: The partial gradients of loss for each parameter of weight is calculated using the chain rule.
4. Updating the Weights : The weights are updated using the following expression [W_{new} = W_{old} - a \cdot K].

Variation of Loss

Loss function vs Epochs @ N=0.1

Loss function vs Epochs @ N=0.2

Loss function vs Epochs @ N=0.5

Loss function vs Epochs @ N=0.8

Loss function vs Epochs @ N=1

Loss function vs Epochs @ N=2

Final Look