About Curriculum learning

[ereday]

Hi,

In Section 5, it is said that :

For each task, we train a separate model, starting with sequences of complexity 6 and incrementing by 4 once it achieves 100% accuracy on held-out examples of the current length. Training stops once the model successfully generalizes to examples of complexity 1000.

From this part, I cannot understand whether you meant the 100% accuracy in training set or test set. Or is help-out examples totally different ? I tried to get the idea from your code but it looks like a bit complicated.