1. Sensory Retina (5×5) S-UNITS
2. Synaptic Weights A-UNITS (W1..W25)
Adjust motorized potentiometers representing synaptic coupling coefficients between sensory retina cells and the summing junction.
Single-Layer Artificial Neuron (Frank Rosenblatt, Cornell Aeronautical Laboratory) vs Multilayer Expansion. Draw patterns on the 5×5 sensory retina, tune potentiometer weights, and inspect linear separability.
Adjust motorized potentiometers representing synaptic coupling coefficients between sensory retina cells and the summing junction.
In July 1958, Cornell psychologist Frank Rosenblatt demonstrated the Perceptron at the U.S. Office of Naval Research. Constructed with 400 photocell retina sensors and motorized potentiometers, the Navy publicly reported it was:
While sensory recognition of letters was demonstrated, Rosenblatt's single-layer machine was fundamentally a linear classifier calculating $y = \text{step}(\sum w_i x_i - \theta)$.
In 1969, Marvin Minsky and Seymour Papert published their mathematical critique Perceptrons, proving that single-layer perceptrons cannot compute non-linear functions like exclusive-OR (XOR) or topological connectedness without multi-layered architectures.
This revelation triggered the first historical AI Winter, halting funding until backpropagation and multi-layer perceptrons (MLP) re-emerged in the 1980s. Today, feedforward blocks inside modern Transformers (such as ChatGPT and Claude) remain direct mathematical descendants of Rosenblatt's layered vision.