Frank Rosenblatt
1928–1971Psychologist and computer scientist who built the first trainable neural network. The Mark I Perceptron was a physical machine that could learn to classify images.
- The Perceptron (1957) — first learning algorithm for neural networks.
- Mark I Perceptron — analog hardware that 'learned' to recognize patterns.
FAQ
What was the perceptron?
A 1957 learning algorithm for a single-layer artificial neuron. Given inputs and labelled examples, the perceptron iteratively adjusts its weights until it can classify them correctly. It was the first concrete, trainable neural-network model and a direct ancestor of every deep network today.
What was the Mark I Perceptron machine?
A 1958 analog computer Rosenblatt built at Cornell Aeronautical Lab. It used a 20×20 grid of photocells as input and a bank of motor-driven potentiometers as adjustable weights — physical hardware that could learn to recognise simple patterns.
Why did perceptron research stall after Rosenblatt?
Minsky and Papert's 1969 Perceptrons proved single-layer networks could not learn XOR. Funding and attention shifted to symbolic AI for nearly two decades, even though multi-layer networks did not share the limitation.
How does Rosenblatt's work connect to modern deep learning?
Every modern neuron is a generalised perceptron: weighted sum of inputs followed by a non-linearity. Deep networks stack perceptron-like units into many layers, and gradient-based training is a direct descendant of the perceptron learning rule.