David Rumelhart

David Everett Rumelhart (June 12, 1942 – March 13, 2011) was an American psychologist who made many contributions to the formal analysis of human cognition, working primarily within the frameworks of mathematical psychology, symbolic artificial intelligence, and parallel distributed processing. He also admired formal linguistic approaches to cognition, and explored the possibility of formulating a formal grammar to capture the structure of stories.

Rumelhart was born in Mitchell, South Dakota on June 12, 1942. His parents were Everett Leroy and Thelma Theora (Ballard) Rumelhart. He began his college education at the University of South Dakota, receiving a B.A. in psychology and mathematics in 1963. He studied mathematical psychology at Stanford University, receiving his Ph.D. in 1967.

From 1967 to 1987, he served on the faculty of the Department of Psychology at the University of California, San Diego. In 1987, he moved to Stanford University, serving as Professor there until 1998.

Rumelhart was elected to the National Academy of Sciences in 1991 and received many prizes, including a MacArthur Fellowship in July 1987, the Warren Medal of the Society of Experimental Psychologists, and the APA Distinguished Scientific Contribution Award. Together with James McClelland, he won the 2002 University of Louisville Grawemeyer Award in Psychology.

Rumelhart became disabled by Pick's disease, a progressive neurodegenerative disease, and at the end of his life lived with his brother in Ann Arbor, Michigan. He died in Chelsea, Michigan. He is survived by two sons.

Rumelhart was the first author of a highly cited paper from 1985 (co-authored by Geoffrey Hinton and Ronald J. Williams) that applied the back-propagation algorithm to multi-layer neural networks. This work showed through experiments that such networks can learn useful internal representations of data. The approach has been widely used for basic cognition researches (e.g., memory, visual recognition) and practical applications. The 1985 paper does not cite earlier publications of backpropagation, such as the 1974 dissertation of Paul Werbos, as they did not know the earlier publications.

Rumelhart developed backpropagation in spring of 1982 independently. In 1983, he showed it to Terry Sejnowski, who tried it and found it to train much faster than Boltzmann machines (developed in 1983). Geoffrey Hinton however did not accept backpropagation, preferring Boltzmann machines, only accepting backpropagation a year later.

In the same year, Rumelhart also published Parallel Distributed Processing: Explorations in the Microstructure of Cognition with James McClelland, which described their creation of computer simulations of perceptrons, giving to computer scientists their first testable models of neural processing, and which is now regarded as a central text in the field of cognitive science.