The first neural network simulation
Farley and Clark simulated a randomly connected network on the Whirlwind machine and showed that it organised itself during training.
Why it matters
A neural network was first run as a program rather than built as a device, and pattern recognition came out of it.
The network held 128 nodes, limited by the machine's memory. The authors observed that after training the system assigned inputs it had never seen to the right class. The follow-up presented in 1955 was titled for exactly that generalisation. The work is often mentioned after the perceptron, though it precedes it by four years.