1928–1971 · psychologist, Cornell

rank Rosenblatt was a psychologist who built a machine that learned. The perceptron is the ancestor, several generations removed, of the neural networks in use today.

what he did

Working at the Cornell Aeronautical Laboratory, Rosenblatt described the perceptron in (Rosenblatt, 1958): a network of simple units that receives input from a "retina", combines it through weighted connections, and adjusts those weights after errors, so that it learns to classify patterns. Building on the logical neurons of McCulloch & Pitts (1943), it added the crucial ingredient: learning from experience.

The idea was first demonstrated as a simulation on an IBM 704, and then built as hardware, the Mark I Perceptron, with a grid of photocells as its eye and adjustable resistors as its weights. A 1958 press demonstration drew newspaper coverage with large claims about what such machines would soon do. Rosenblatt set out the theory at length in (Rosenblatt, 1962).

what happened next

Minsky & Papert (1969) proved that a single-layer perceptron cannot learn some simple functions, such as exclusive-or, and their analysis is often credited with turning funding and attention away from neural networks for years. Multi-layer networks, which can learn such functions, became practical once a good training method for them was widely adopted in the 1980s. Rosenblatt died in a boating accident in 1971 and did not see it.

Primary works and sources

1. Rosenblatt (1958). The perceptron: A probabilistic model for information storage and organization in the brain. Psychological Review. doi:10.1037/h0042519
2. McCulloch & Pitts (1943). A logical calculus of the ideas immanent in nervous activity. The Bulletin of Mathematical Biophysics. doi:10.1007/BF02478259
3. Rosenblatt (1962). Principles of Neurodynamics. Washington, DC: Spartan Books.
4. Minsky & Papert (1969). Perceptrons. Cambridge, MA: MIT Press.

see also

	timeline: machines that answer — 1958
	shortcut learning — what a learning machine can learn by mistake
	swarms — simple units, collective behavior
	Donald Hebb — the learning rule the perceptron borrowed from