Short answers, pulled from the story.
Kunihiko Fukushima introduced the multilayer visual feature detection network called the neocognitron in 1969. This design featured elements where all units in one layer shared identical interconnecting coefficients across homogeneous arrangements.
LeNet-5 began fielding in American banks since June 1996 after emerging as a pioneering seven-level network classifying hand-written numbers on checks digitized into 32 by 32 pixel images in 1995.
Standard neural networks could run twenty times faster on graphics processors starting in 2004 when K.S. Oh and K. Jung showed this capability compared to equivalent CPU implementations.
Facial recognition systems achieved ninety-seven point six percent accuracy on five thousand six hundred still images while achieving large decreases in error rates with one paper reporting these results.
The AtomNet system introduced in 2015 predicted novel candidate biomolecules for diseases like Ebola virus and multiple sclerosis through three-dimensional chemical representations of molecule-biological protein interactions.