Perceptron and artificial neural networks

Rosenblatt's perceptron was the first machine that learned from examples by adjusting weights, the ancestor of every modern neural network.

AI & roboticsSoftware
  1. 1958

    First version

    Frank Rosenblatt (Cornell Aeronautical Laboratory)

    Image: National Museum of the U.S. Navy · Public domain · Wikimedia Commons
  2. 1980

    Neocognitron (Fukushima)

    Hierarchical, convolution-like network for shift-invariant pattern recognition.

    Image: Japan Broadcasting Corporation · Public domain · Wikimedia Commons
  3. 1982

    Hopfield network

    Recurrent network stores patterns as stable energy states, recalling whole memories from fragments.

    Image: Saul Bass · Public domain · Wikimedia Commons
  4. 1989
    LeNet convolutional network (LeCun, Bell Labs)

    LeNet convolutional network (LeCun, Bell Labs)

    Backprop-trained CNN reads handwritten ZIP-code digits.

    Image: Pauli Rautakorpi · CC BY 3.0 · Wikimedia Commons

Contested: McCulloch and Pitts' 1943 artificial neuron and Hebb's 1949 learning rule preceded it; Rosenblatt built the first learning implementation.

See it on the timeline