Image recognition works in a similar way, where instead of taking chunks of text, we take segments of images.
AI image models break models down into segments
We must remember that each image is just a collection of numbers representing red, blue and green (RGB). The AI learns what combinations of pixels in what position represents each group of images. If we randomly change some colours in the image, let’s see what happens
Introducing noise can greatly vary the output of a model
Although we see the image as a cat, since the RGB values of the image have changed the AI recognises it as something different. This scenario is another representation that AI is ultimately a very large set of mathematical models.