In the notebook Course 1 - Part 6 - Lesson 2 - Notebook the accuracy of a feed forward is being compared to a CNN with a claim that the train accuracy went from 89% to 93% and the test accuracy 87% to 91%.
However, this appears to be an apple to oranges comparison as the number of trainable parameters in the CNN model is more than twice that in the feed forward.
Ideally two models with same depth and same number of trainable parameters ought to be compared.
In the notebook Course 1 - Part 6 - Lesson 2 - Notebook the accuracy of a feed forward is being compared to a CNN with a claim that the train accuracy went from 89% to 93% and the test accuracy 87% to 91%.
However, this appears to be an apple to oranges comparison as the number of trainable parameters in the CNN model is more than twice that in the feed forward.
Ideally two models with same depth and same number of trainable parameters ought to be compared.