Hi all,
We have an AI lab Pulseinnovas and our AI lab has discovered an new architecture which beats Google’s Transformer architecture…Google’s transformer architecture took 20k epochs of training to generalize over benchmark dataset…while our model just did it at 650 epochs thereby cutting the training cost by 98-99…i have attached our benchmark run results
The blue line attending the level of accuracy is Transformer architecture by Google and orange line is our model on X axis we have number of epochs and y axis we have test accuracy
