I thought that ML might stall on object recognition, based on what I knew about how it works in humans. Seeing that prediction fail didn't make me think "we'll never hit a wall on anything" but did make me much more reticent to say "big data + scaled up models can't achieve X".https://twitter.com/jachiam0/status/1153183048667561984 …
I had exactly the same response. The ImageNet (& followon) results shocked me in a way no prior result in ML/AI had.
-
-
It is very impressive. But also still far below humans on object recognition / 1
-
The networks that do well on ImageNet are somewhat brittle -- when images are perturbed (not just adversarial perturbations, but also random noise) or objects rotated, or other "out of sample" changes applied, the networks fail even though humans are very robust. /2
- 4 more replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.