uoaei on NoProp: Training neural networks without back-propagation or forward-propagation · Hackerly

Comment on NoProp: Training neural networks without back-propagation or forward-propagation parent

And there's good reasons why we use gradients today.