Great Moments in Image Recognition: Rifle or Turtle?
November 7, 2017
I read “AI Image Recognition Fooled by Single Pixel Change.” The write up explains:
In their research, Su Jiawei and colleagues at Kyushu University made tiny changes to lots of pictures that were then analyzed by widely used AI-based image recognition systems…The researchers found that changing one pixel in about 74% of the test images made the neural nets wrongly label what they saw. Some errors were near misses, such as a cat being mistaken for a dog, but others, including labeling a stealth bomber a dog, were far wider of the mark.
Let’s assume that these experts are correct. My thought is that neural networks may need a bit of tweaking.
What about facial recognition? I don’t want to elicit the ire of Xooglers, Apple iPhone X users, or the savvy folks at universities honking the neural network horns. Absolutely not. My goodness. What if I at age 74 wanted to apply via LinkedIn and its smart software for a 9 to 5 job sweeping floors?
Years ago I prepared a series of lectures pointing out how widely used algorithms were vulnerable to directed flows of shaped data. Exciting stuff.
The write up explains that the mavens are baffled:
There is certainly something strange and interesting going on here, we just don’t know exactly what it is yet.
May I suggest an assumption that methods work as sci fi and tech cheerleaders say they do is incorrect?
Stephen E Arnold, November 7, 2017
Comments
-
- Subscribe to Beyond Search
Feature archive
News archive
-
