The first NIPS conference

In November 1987 about six hundred people met in Denver for the first Neural Information Processing Systems conference, organised largely by Terrence Sejnowski and held in the mountains so that the physicists, engineers and neuroscientists who attended could ski together afterwards. The mix was the point. Backpropagation had made networks interesting to computer scientists; Hopfield had made them interesting to physicists; the biologists had been there all along. NIPS gave them a common venue at the moment the rest of AI was heading into its second winter.

For the next two decades the conference stayed small and somewhat marginal, a place where kernel methods, Bayesian statistics and the stubborn remnant of neural-network research argued with each other while the mainstream AI conferences did logic and planning. Its proceedings from those years hold most of the ideas the deep-learning era was built on.

After 2012 it became the centre of the field. Attendance passed 8,000 in 2018 and tickets sold out in minutes; the conference changed its name to NeurIPS that year. Every model on this timeline from AlexNet onwards was either presented there or measured against something that was.