CIFAR launches the Neural Computation & Adaptive Perception program

Background

Geoffrey Hinton first joined the CIFAR community in 1987, as a Fellow in the institute's first program, then called Artificial Intelligence, Robotics & Society. He kept working on neural networks through a long period in which most of the field was skeptical of the approach and funding for it was scarce ^[1].

What happened

In 2004, Hinton and collaborators — among them Yoshua Bengio and David Fleet — successfully proposed the launch of a new CIFAR research program called Neural Computation and Adaptive Perception, usually shortened to NCAP ^[1]. CIFAR itself describes the support it gave that year as the spark that set off the deep learning era, and describes NCAP as the predecessor of the program that exists today under the name Learning in Machines & Brains ^[2].

Key details

The proposal set out two objectives. The first was to draw on insights from a broad range of disciplines in order to understand how the brain learns and interprets visual information. The second was to build computer systems and machines that could learn in ways similar to humans ^[1]. Bridging neuroscience and machine learning was therefore part of the design from the start, rather than something added later.

Hinton went on to lead NCAP for ten years ^[1]. The early membership named by CIFAR reads like a list of the people who would later define the field: Yoshua Bengio, Yann LeCun, Andrew Ng, Sebastian Seung, Max Welling and Ruslan Salakhutdinov ^[1]. Bringing them into a single, funded, long-running program gave a scattered group of researchers a shared institutional home.

Training was part of the program from early on. By 2005 NCAP had launched its first summer school, then called the Neural Computation & Adaptive Perception Summer School, held in Toronto with 28 competitively selected participants ^[1].

Impact

The summer school outlived its original name and format. It is now known as the Deep Learning + Reinforcement Learning Summer School, and CIFAR reports that it has grown to more than 2,700 alumni ^[1]. That makes it one of the routes through which a program founded around a contested idea turned into a regular channel for training new researchers.

The people involved also went on to receive the field's highest recognition. Yoshua Bengio and Yann LeCun would later share with Hinton the 2018 ACM A.M. Turing Award, given for their development of deep learning ^[1]. What began in 2004 as a proposal for a single CIFAR program thus connects the era when neural networks were a minority pursuit to the era in which they became the basis of mainstream machine learning.