← Back to Computer vision
All 10 questions. After you submit, each question shows the right answer and why. Back to the quick check
A computer can be tricked into getting a picture wrong by changes so tiny that a person can't even see them.
In a deep network that recognises pictures, what do the first layers usually pick out?
What is a convolutional filter's job when it slides across an image?
In the 2012 ImageNet competition, what top-5 error did the winning deep convolutional network achieve?
A 3×3 filter has columns −1, 0, +1. It sits on a patch whose left two columns are 0 and whose right column is 9. What is the filter's output?
In the 2018 Gender Shades audit, what was the highest error rate (in %) found for darker-skinned females?
Which three ideas did LeCun and colleagues say convolutional networks combine?
What did LeNet-5's MNIST test error fall to when it was trained with artificially distorted examples?
The ImageNet analysis showed that every human tester was more accurate than the best model of the time.
What did Szegedy et al. (2013) find about whether an adversarial perturbation transfers between networks, and why does that matter for deployed vision systems?
