David E. Rumelhart (1942-2011) was a mathematical and cognitive psychologist whose work helped revive neural networks in the 1980s. He earned a PhD in mathematical psychology from Stanford in 1967 and taught at UC San Diego from 1967 to 1987 and then at Stanford until 1998. The Cognitive Science Society’s biography credits him with developing models across cognition “from motor control to story understanding to visual letter recognition to metaphor and analogy.”
Rumelhart co-led the parallel distributed processing (PDP) group, whose 1986 two-volume book argued that the mind’s abilities emerge from networks of simple, neuron-like units adjusting their connections. According to the society, “he formulated the powerful back-propagation learning algorithm for training networks of neuron-like processing units,” and the 1986 Nature paper he co-authored with Geoffrey Hinton and Ronald Williams brought that algorithm to wide attention. Backpropagation remains the core training method for essentially every deep neural network today.
Rumelhart was elected to the National Academy of Sciences in 1991 and received a MacArthur Fellowship and the APA Distinguished Scientific Contribution Award. He died in 2011 from Pick’s disease, and the field’s leading award in cognitive science, the Rumelhart Prize, is named in his honor.