David Rumelhart
1942–2011 · American psychologist and cognitive scientist
David Everett Rumelhart (1942-2011) was an American psychologist and cognitive scientist whose work helped make formal models a central tool for explaining the human mind. Trained in mathematical psychology, a field that describes mental processes through equations and simulations, he worked at the boundary between psychology, linguistics and artificial intelligence.
He began his studies at the University of South Dakota and earned his doctorate at Stanford University in 1967. From 1967 to 1987 he taught in the psychology department at the University of California, San Diego, before moving to Stanford. His early research examined how people understand stories and how mental schemata, structures of knowledge that organize information, shape comprehension and memory.
His most influential work was Parallel Distributed Processing (1986), written with James McClelland, which argued that cognition emerges from information processed simultaneously across networks of neuron-like units rather than from a single linear sequence of operations. That same year, with Geoffrey Hinton and Ronald Williams, he presented the back-propagation algorithm, which allows multi-layer networks to adjust their connection weights. He died in 2011 after living with Pick's disease, a degenerative form of dementia.
His influence persists in deep learning and in much of today's artificial intelligence.
Key ideas
- Parallel distributed processing: information is handled at once by many connected nodes, much as in the brain, rather than through a linear chain of steps.
- Mental schemata: the mind organizes what it perceives into prior structures that simplify understanding and recall.
- Back-propagation: a network adjusts the weights of its connections to reduce output error, which lets it learn complex patterns.
Curiosities
A method developed independently
According to the history of the algorithm, Rumelhart developed it independently in 1982, and the 1986 paper did not cite earlier work, such as Paul Werbos's 1974 dissertation.
An early test
In 1983 he showed the method to Terry Sejnowski, who found that it trained networks much faster than Boltzmann machines.
Original Filopedia text based on Wikipedia (es) and Wikipedia (en) and Wikidata.