tensorflow - Overfitting DL model?

Question

I am trying to build a Deep Learning model to pick out Tropical Cyclones in weather model data. I have collected the data, normalized it in the region [0, 1] and passed it to my early model. Then I plotted my loss and accuracy curves as here. I am getting weird curves as the validation loss starts increasing after ~50 epochs, indicating overfitting, but the validation accuracy is still increasing. Is my model overfitting (at around 50 epochs) or not?

标签： tensorflowkerasdeep-learning

Answer 1

These graphs are classical graphs that comes with overfitting! You can recognize overfit because even though training accuracy keeps increasing, validation accuracy does not. To prevent overfit, there are several approaches; too numerous to name in one answer. You could apply L1/L2 regularization, dropout, or artificially expand your training data (amongst others).