How Does a Neural Network Learn?
From a single artificial neuron to a network that learns
Loading…
Summary
An artificial neuron is a simple unit of calculation that multiplies its inputs by weights, adds them up and says "yes" or "no" depending on the result. Changing the weights changes what the neuron says "yes" to. Learning means adjusting the weights step by step in the direction that reduces the error. Connect neurons in layers and the network can tell apart complex patterns that a single neuron could never separate.
In depth
Warren McCulloch and Walter Pitts proposed the first model of an artificial neuron in 1943; in 1958 Frank Rosenblatt turned it into a machine that could learn, the perceptron. In 1969 Marvin Minsky and Seymour Papert showed that a single-layer network cannot learn even a simple pattern such as XOR. The solution was to arrange neurons in layers, together with backpropagation, a method that works out each weight's share of the error; it became widespread with the 1986 paper by Rumelhart, Hinton and Williams. After 2012, with big data and powerful graphics processors, "deep" networks became the foundation of artificial intelligence, from image recognition to language models. The 2024 Nobel Prize in Physics was awarded to John Hopfield and Geoffrey Hinton for pioneering work in this field.
Rate How Does a Neural Network Learn?
What did you think of this content? Click the stars to leave your rating.
Leave a Comment and Rating
Sign in to comment and rate. Membership is free.