Why the validation error does not decrease with a training of 30 epochs but decreases with a training of 60 epochs When i train my model with 30 epochs, the training and validation error curves seems to stagnate:

However, when i train my model with 60 epochs, the training and validation error curves start to decrease:

Can anyone explain me why? In the first graph, shouldn't the minimum validation error be closer to 0.10 instead of 0.15?
Thank you!
 A: It is difficult to say what happened when we see only the images and there is no explanation how you made these images.
There are multiple possibilities:

*

*You have some algorithm that changes when you run 60 epochs instead of 30 epochs. For instance the stepsize at the start might be different and this could change the convergence path. (This option seems unlikely, I believe that not many algorithms would change their behaviour when you plan to run it a different number of epochs)


*The difference might occur if your method has some randomness and the second time that you did the training with the different number of epochs you got a different curve that was closer to 0.10 at 30 epochs because of that randomness (and not because of the epochs that you plan to run it).
The randomness can occur because of different reasons. It can be that the splitting of the data into training and validation is random and was different the second time. It can be that the algorithm uses random steps, e.g. stochastic gradient descent, or uses a random starting value, and because of that the convergence might be different and in the first case the algorithm got stuck in a local minimum.
