Loading slide
Loading contents...
Rosenblatt's first Perceptron ran as a program on an IBM computer. Later, his team built the idea into a physical machine called the Mark I Perceptron.
It came in two sections, each nearly two metres tall. Together, they filled more than three metres of wall.
A pattern was shown to 400 light-sensitive cells. Each cell turned one small patch of light or dark into an electrical input.
The signals travelled through fixed wiring into a bank of adjustable electrical dials. Their positions stored the weights. A response panel showed the machine's answer.
When the answer was wrong, the learning procedure sent back a correction signal. Small electric motors turned the dials, changing the weights before the next example.
The trust-dial analogy was literal. In the Mark I, learning moved physical dials.