Why do we need backpropagation?

Does an approach exist by which we train without back-propagation.
How to remove back-propagation ?

it doesnt make sense to train a model without backpropagation because the model has to adjust the weights and bias to make the model perform better and better.

As of my knowledge there is no other approach than doing backpropagation

There are some networks called capsule networks which uses routing by agreement but I still have to fully read those papers :slight_smile:

Neuroevolution has proven to be a good alternative in some cases.