SEER Suckers

Facebook uses Instagram images to train AI recognition engine called SEER.

Apparently over a billion publicly accessible images were sent to the algo, stripped of any labelling and then processed. The results were almost 85% accurate which is pretty good bearing in mind the images could have been of anything. However, it wouldn’t surprise me to learn that the majority of images were faces.

The algo attempts to attribute each pixel to the object to be recognised. So in a farm setting it should be able to recognise cows from trees. I doubt it would do that well with some of Pablo Picasso’s paintings though.

SEER is an abbreviation of self-supervised, so one assumes no human intervention other than to corroborate the images were identified correctly (or not).

For those interested, seersucker is a fabric which has a striped, puckered appearance caused by warp threads being fed at a greater rate on alternate stripes. In its use as the title, its being a little playful with words.

Leave a Comment