Sudden drop in accuracy

In week3 assignment 2 ( image segmentation with u- net) , I change the default epochs =5 to 40. Then i got accuracy over epochs plot as following with sudden dorp in the accuracy. What could be reason for that.

It could be that when you are training, you are not using all the training data (depending on steps per epoch) and by the 27th epoch depending on the shuffling of the data it gets fed images it wasn’t fed before…

The solution surfaces are incredibly complicated. There is never any guarantee of monotonic convergence: you can “go off a cliff” at any iteration.

If this assignment uses batch training, it could be that at the 27th iteration, you got a batch that has new data which the model had not yet been trained to recognize.