AIncorrect
Overfitting
Overfitting is a state where accuracy is high on the training data but low on unseen data.
It differs from this scenario, where accuracy is also low on the training data, so it is incorrect.
An ML team validated a classification model it had built and found that accuracy remained low on both the training data and the test data. The model does not seem to capture the patterns in the data well. What is this state called?
1 / 1
Select an answer
CorrectB
Explanation
A question on naming the state with low accuracy on both training and test data.
- 1「accuracy remained low on both the training data and the test data」It fails on either one = underfitting
BCorrect
Underfitting
Correct. Underfitting is a state where the model is too simple or undertrained, so it cannot even fit the training data well, and accuracy is low on both the training and test data.
CIncorrect
Data drift
Data drift is a phenomenon where accuracy drops because the input distribution changes during operation.
This scenario shows no accuracy on any data from the training stage onward, which is not a distribution-change problem, so it is incorrect.
DIncorrect
Vanishing gradients
Vanishing gradients is a technical problem where the update signal does not propagate in deep neural networks.
It can be one cause of failed training, but the name for the state of 'low accuracy on any data' is underfitting, so it is incorrect.
Key Takeaway
Grasp the correct answer, 'underfitting (insufficient learning)'.
・A state where the model is too simple or undertrained, so it cannot even fit the training data well.
・As a result, accuracy is low on both the training and test data.
Overfitting is a different state with high accuracy on training and low on unseen data, while data augmentation (a technique) and inference (a process) are not names of a 'state' at all.