So the best part of backwards prop is that you need very little memory to complete the training which matters on GBs of data. FF seems to allow for more paralle

So the best part of backwards prop is that you need very little memory to complete the training which matters on GBs of data. FF seems to allow for more parallel computing but you will need to keep at least one more set of coefficients more in storage to deal with scheduling conflicts? Definitely merits more work.