Keeping up my trend this week of finding out well known illustrative stories are most likely false, I found @gwern β€˜s debunking of the AI rank story gwern.net/Tanks

The Neural Net Tank Urban Legend

AI folklore tells a story about a neural network trained to detect tanks which instead learned to detect time of day; investigating, this probably never happened.

gwern.net
4
25
2
118
Replying to @goodfellow_ian @gwern
We made exactly the same mistake in one of my projects on insect recognition. We photographed 54 classes of. Insects. Specimens had been collected, identified, and placed in vials. 1/

7:40 PM Β· Jul 26, 2019

2
4
2
23
Replying to @tdietterich @goodfellow_ian @gwern
Vials were placed in boxes sorted by class. I hired student workers to photograph the specimens. Naturally they did this one box at a time; hence, one class at a time. 2/
1
1
0
8
Photos were taken in alcohol. Bubbles would form in the alcohol. Different bubbles on different days. The learned classifier was surprisingly good. But a saliency map revealed that it was reading the bubble patterns and ignoring the specimens. 3/
1
1
0
26
more replies
Replying to @tdietterich @goodfellow_ian @gwern
This is basically batch effect. molmine.com/magma/global_ana…
0
0
0
0