Practice quiz: Methods and results

4 questions. Pick an answer to see if it is right. You can try as many times as you like. These are for practice: the final test has different questions.

1. 1.Settings were tuned using the test set. What does this do to the reported result?

   Makes it more reliableMakes it too optimisticHas no effectMakes it pessimistic
2. 2.What did Melis, Dyer and Blunsom (2018) find?

   LSTMs cannot be tunedWell-tuned older LSTMs beat several newer models that claimed to be betterNewer models were always betterTuning never matters
3. 3.An ablation study...

   adds more dataremoves one part of a method at a time to see what each part contributestests on a new datasetcompares two papers
4. 4.In the ResNet paper, what happened to a plain network when it went from 18 to 34 layers?

   Training error went downTraining error went upNothing changedIt could not be trained