Overfitting to the dev set?

i don’t understand how the model can overfit to the dev set. we only use it to measure the performance of the model. we don’t apply backpropagation or update the weights of the model so how could that be possible?

If you adjust the training regularization using feedback from the dev set cost, then you could overfit the dev set.

Hi @Omar_Mostafa1,

Would you mind to share the source of that statement - overfit to the dev set? I am interested in that too.