Frank Rosenblatt
Psychologist who built the perceptron in 1958, the first neural network that learned its own weights from data rather than having them set by hand.
Frank Rosenblatt (1928–1971) took the McCulloch-Pitts neuron — a fixed logic gate with hand-set weights — and gave it a learning rule. His perceptronPerceptronThe Perceptron (1958) was the first learning system built from an artificial neuron, and the direct ancestor of the neural network training loop., built in 1958 as both an algorithm and a physical machine (the Mark I Perceptron, wired to a 20×20 grid of photocells), adjusted its own weights based on whether it classified an example correctly. Get it wrong, nudge the weights toward the right answer; repeat over the training set until it converges. That update rule is a direct, if simpler, ancestor of gradient descentGradient DescentGradient descent is the optimization algorithm that trains models by repeatedly stepping parameters in the opposite direction of the loss function's gradient. and backpropagationBackpropagationBackpropagation is the algorithm that computes the gradient of a neural network's loss with respect to every parameter, by applying the chain rule backward through the network..
Rosenblatt was also an aggressive promoter of his own work, and press coverage at the time suggested perceptrons would soon see, speak, and think — expectations the single-layer perceptron could never meet, since it can only separate data that's linearly separable. When Marvin MinskyMarvin MinskyCo-founder of the MIT AI Lab and a founding figure of the field — his 1969 book with Seymour Papert exposed the perceptron's limits and helped trigger the first AI winter. and Seymour Papert's Perceptrons demonstrated this limitation rigorously in 1969, the gap between the hype and the math became impossible to ignore, and funding for neural network research collapsed for most of a decade.
Rosenblatt died in a boating accident in 1971, before the field revisited his ideas with the multi-layer networks and backpropagationBackpropagationBackpropagation is the algorithm that computes the gradient of a neural network's loss with respect to every parameter, by applying the chain rule backward through the network. that fixed exactly the limitation his critics had identified.
See also: Marvin MinskyMarvin MinskyCo-founder of the MIT AI Lab and a founding figure of the field — his 1969 book with Seymour Papert exposed the perceptron's limits and helped trigger the first AI winter., Warren McCullochWarren McCullochNeurophysiologist who, with Walter Pitts, proposed the first mathematical model of a neuron in 1943 — the ancestor of every neural network built since.
Learn more: Neural Networks & Backprop · Wikipedia: Frank Rosenblatt
Mentioned in
Lessons where this comes up in context.
Marvin Minsky
Co-founder of the MIT AI Lab and a founding figure of the field — his 1969 book with Seymour Papert exposed the perceptron's limits and helped trigger the first AI winter.
David Rumelhart
Cognitive scientist who, with Geoffrey Hinton and Ronald Williams, popularized backpropagation in 1986 — the algorithm that made multi-layer neural networks trainable.