Training data quiz

All 8 questions. After you submit, each question shows the right answer and why. Back to the quick check

ExplorerQuestion 1 of 8

If most of the photos a computer learns from show one kind of face, it can end up worse at recognising other kinds of face.

True

False

ExplorerQuestion 2 of 8

Why do people keep some examples back as a 'test set' instead of using them all for training?

Because computers can only hold a few examples

Because test examples are always harder

To make training slower

To check how well the model does on examples it has never seen

InvestigatorQuestion 3 of 8

A 2021 study checked the test sets of 10 popular datasets for wrongly labelled examples. What did it find?

On average at least 3.3% of test labels were wrong, and at least 6% in the ImageNet validation set

No errors at all, because experts made every label

Exactly half of all labels were wrong

Errors only in audio datasets

InvestigatorQuestion 4 of 8

How many training examples does the MNIST handwritten-digit database contain?

Your answer

ScientistQuestion 5 of 8

According to NIST, bias in an AI system can only arise if someone involved intended to discriminate.

True

False

AdvancedQuestion 6 of 8

A model was trained on 2015 shopping data and is deployed in 2026. Which NIST-identified risk is most directly at play?

The model being too small to store 2015

The model running out of memory

Training data becoming stale or detached from the deployment context

AdvancedQuestion 7 of 8

A language model scores highly on a quiz whose questions and answers were already posted on many websites before the model was trained. Which problem should you suspect?

Too small a learning rate

Data contamination

Underfitting

Label noise in the training set

ExpertQuestion 8 of 8

Name four things that Gebru and colleagues proposed a dataset's datasheet should document, and the engineering practice that inspired the idea.

Your answer

Methodology
Sources
Safety
Editorial policy
Privacy
Terms
Cookies
Contact
Accessibility
Site map