Neural networks

Back propogation. 

As with gradient descent, the idea is simple:

1. find Unknown macro: formula. Click on this message for details.
   The [formula] macro is not in the list of registered macros. Verify the spelling or contact your administrator. for each parameter in neural network (weights, biases)
2. change weights to have better J (minimize)

Can we start from the end (final total error)? 

We have to understand how each particular weight impact the final objective function J passing through the whole chain.

Back propogation derivation. 

1. Change in objectif function by shifting a_k^{L-1} - activation in layer L-1 for neuron k.

2. neuron error in L-1 as function of neuron errir in L

Recall Unknown macro: formula. Click on this message for details.
The [formula] macro is not in the list of registered macros. Verify the spelling or contact your administrator.