フリー問題

問題文

A team sees the loss jump up and down without settling. The same model and the same data train smoothly once one single number is made ten times smaller. Which setting was too large?

選択肢

1. The setting that was too large is regularization strength of the optimizer.
2. The setting that was too large is layer count of the optimizer.
3. The setting that was too large is batch capacity of the optimizer.
4. The setting that was too large is learning rate of the optimizer.