In spite of applying our greatest efforts at planning and coaching the neural network, typically a specific network merely won’t converge on an answer that's acceptable to the system necessities.
It could get close, but not meet our requirements.
Now want to know how Network Convergence Fails
This outcome will happen as a result of there aren’t enough nodes to remodel the computer file into correct outputs.
Perhaps you don’t have enough coaching knowledge, or the coaching knowledge wasn’t collected with knowledge integrity in mind.
Here are 37 Reasons why your Neural Network is not working.
You can try the following steps, as to resolve your problem occurred in the back prop neural network :
Implemented momentum (and kept the value at 0.5)
Kept the learning rate at 0.1
Charted the error, weights, input as well as output of each and every neuron, Seeing the data as a graph is more helpful in figuring out what is going wrong
Tried out different activation functions (all sigmoid). But this did not help me much.
Initialized all weights to random values between -1 and 1.
Since you were implementing a neural network from scratch, it turned out that there was an error in the update function. You can found your error through gradient checking.