On Wed, Dec 19, 2018 at 01:16:38PM +1300, Lawrence D'Oliveiro wrote:
Some good background on the origins of all the current hubbub over automated image recognition: <https://arstechnica.com/science/2018/12/how-computers-got-shockingly-good-at-recognizing-images/>. A key breakthrough came in 2012, when a team at the University of Toronto combined the “convolutional” neural network concept with a massive deployment of sheer computing power to leave its competitors in the dust in the “ImageNet” image-recognition competition that year.
That team are now working at Google.
Oh, and the recognition algorithms are now scoring better than many humans in some areas.
On images like what they were trained on. They can completely fail on simple (sometime invisible) modifications to images which humans are not fooled by. Cheers Michael.