Glossary
Who is Frank Rosenblatt?
In short
Frank Rosenblatt was a psychologist and engineer who in 1958 described the perceptron, one of the first neural networks that could learn from examples.
People · Frank Rosenblatt
MMOOZZGG.com editorsChecked: 2 min read
For a child
Imagine you are teaching your little brother to tell apples from pears. He makes a mistake, you say, “No, that’s a pear”, and next time he looks more carefully. Frank Rosenblatt built a machine that learned the same way: it looked at pictures, got a hint when it was wrong, and slowly corrected itself. It was one of the first machines that learned on its own instead of only following ready commands.
In plain words
Rosenblatt worked as a psychologist and engineer at the Cornell Aeronautical Laboratory. In 1958, he described the perceptron, a model of a simplified nervous system. The perceptron receives signals, adds them up with different weights and answers “yes” or “no”. If the answer is wrong, the weights shift a little, and over time there are fewer errors. In July of the same year, the US Office of Naval Research showed an experiment: after 50 tries, an IBM 704 computer learned to tell cards marked on the left from cards marked on the right. Newspapers wrote about a machine that learns by itself, although nobody yet knew how to train multilayer networks.
For business
Rosenblatt’s perceptron stands at the start of the family tree of almost every model that sorts things in a company today: requests by department, emails into useful and spam, payments into normal and suspicious. The principle is the same: the model sees features, weighs them and picks a class, and errors on labeled examples adjust the weights. A manager can take three lessons from this. You need examples with correct answers, or there is nothing to learn from. A simple model learns fast and runs cheaply, but it misses complex dependencies. And big promises are worth checking: in 1958, Rosenblatt wrote about the near birth of a machine that perceives and recognizes its surroundings, but decades passed before real use. Sorting tasks are a convenient place to start automating request handling.
For a specialist
The paper “The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain” appeared in Psychological Review (vol. 65, no. 6, pp. 386–408) in 1958. Rosenblatt builds a theory of a hypothetical nervous system that serves as a bridge between biophysics and psychology and makes it possible to predict learning curves. In modern terms, the perceptron is a linear threshold unit with the perceptron learning rule: on an error, the weights shift by the input value with the sign of the error. For linearly separable classes, the algorithm converges in a finite number of steps. The hardware Mark I Perceptron worked as a classifier of visual patterns, and the next machine, Tobermory, was meant for speech recognition. In 1959, Rosenblatt became head of the cognitive systems program at Cornell. The book “Perceptrons” by Minsky and Papert (1969) showed the limits of the single-layer design, and the answer came with multilayer networks trained by backpropagation.
Example
A delivery service trains a simple model to tell urgent messages from regular ones by the words in the text. The model makes mistakes, an operator corrects the label, the weights shift as in Rosenblatt’s perceptron, and after a few weeks there are clearly fewer errors.
Related terms
Questions and answers
Sources
Where companies use it
Not sure whether your company needs this?
AI solutions map