The Neural Lattice

Data, GPUs and old ideas finally meet.

Deep belief nets in 2006 revived interest in many-layered networks. ImageNet supplied millions of labelled images; in 2012 AlexNet, trained on two gaming GPUs, cut the error rate so sharply that computer vision changed overnight.

Within four years neural networks were translating, generating faces that never existed and, in March 2016, playing a move at Go that professional commentators first took for a mistake.

Events

See it in the chronicle