2009
ImageNet
14 million labelled images — the fuel without which AlexNet could not exist.
Milestones on the line · PERCEPTION
Deep learning
Computers learned to tell what is in a picture from examples rather than from hand-written rules — the way a child does.
Convolutional nets beat humans at image recognition. Machines learned to see — the starting point of the modern AI wave.
StatusPASSED
TypeMilestones on the line
Marker2012
What exists today7
Researched
2012
AlexNet
ImageNet error collapses — the GPU training era begins.
2014
Adversarial networks
GANs: machines start convincingly generating, not just recognising.
2015
ResNet
Residual connections make hundred-layer networks trainable.
2016
AlphaGo
Beats Lee Sedol: search plus reinforcement learning.
In progress
сейчас
Vision as a component
Recognition stopped being a product of its own and became a built-in function of large models.
Distant horizons
впереди
Perception without labels
Learning from raw camera and sensor streams with no labelled examples at all.
Milestones on the line
To the line