Warning: This description contains the answer key, so do not read this description unless you would like to see the solutions to the 20 test images.
In this video, we turn the MNIST training and test data set into a challenge that is difficult for humans to overcome. In this video, we give humans examples of the magnitudes of Fourier transforms of MNIST digits (these are handwritten digits in the set {0,...,9}). After exposure to the magnitudes of the Fourier transforms, we ask if the participants can recognize the digits based on the magnitudes of the Fourier transforms alone.
After obtaining the absolute value images of the handwritten digits, we do two image processing steps to make the images more clear. We first remove the (1,1) entry in the Fourier transforms of the images because the (1,1) entry has the highest magnitude and removing this entry will make the rest of the image brighter. To further brighten up the image, we take the square roots of the magnitudes of the Fourier transform.
The notion of the Fourier transform is not my own. I just made this animation since the magnitude of the Fourier transform is translation invariant and translation invariance is useful for image processing.
Caution: Solutions to the test items are given below. Do not look below here unless you really mean to do so.
.
.
.
.
.
.
.
.
.
.
Test item 1 : answer 9
Test item 2 : answer 5
Test item 3 : answer 2
Test item 4 : answer 8
Test item 5 : answer 3
Test item 6 : answer 9
Test item 7 : answer 0
Test item 8 : answer 3
Test item 9 : answer 9
Test item 10 : answer 9
Test item 11 : answer 6
Test item 12 : answer 6
Test item 13 : answer 0
Test item 14 : answer 2
Test item 15 : answer 3
Test item 16 : answer 1
Test item 17 : answer 2
Test item 18 : answer 8
Test item 19 : answer 9
Test item 20 : answer 3
Unless otherwise stated, all algorithms featured on this channel are my own. You can go to github.com/sponsors/jvanname to support my research on machine learning algorithms. I am also available to consult on the use of safe and interpretable AI for your business. I am designing machine learning algorithms for AI safety such as LSRDRs. In particular, my algorithms are designed to be more predictable and understandable to humans than other machine learning algorithms, and my algorithms can be used to interpret more complex AI systems such as neural networks. With more understandable AI, we can ensure that AI systems will be used responsibly and that we will avoid catastrophic AI scenarios. There is currently nobody else who is working on LSRDRs, so your support will ensure a unique approach to AI safety.