Universal topology neural network
Posted: Sun Nov 29, 2020 8:06 pm
Hi all!
I've made my first neural network, with my own topology and activation function.
This implementation searches the answer for XOR, it needs just 5 nodes:
2 input nodes;
2 middle (hidden nodes)
1 output node.
Every nod has own "layer", but every node is connected to all previous nodes. So, for this 5 nodes we are need 10 weights, 5 biases.
Actually, first two biases are not needed, first weight too.
The graph shows the error for all 4 examples.
There is just only small mutation, I need to fix it.
I've made my first neural network, with my own topology and activation function.
This implementation searches the answer for XOR, it needs just 5 nodes:
2 input nodes;
2 middle (hidden nodes)
1 output node.
Every nod has own "layer", but every node is connected to all previous nodes. So, for this 5 nodes we are need 10 weights, 5 biases.
Actually, first two biases are not needed, first weight too.
The graph shows the error for all 4 examples.
There is just only small mutation, I need to fix it.