Teuvo Kohonen, Kaoru Nakano and James Anderson Publish Associative Memory Networks

Neural network research needed a win. It got that and began its long road to recovery after the Perceptrons book setback, especially in the West, with Associative Memory networks.

In 1972 three researchers working separately published network models of associative memory: Teuvo Kohonen in Finland with a “correlation matrix memory,” Kaoru Nakano with the Associatron and James A. Anderson of Rockefeller University in New York with an “interactive memory.” Each stored many patterns across the same set of connections, so that a whole item could be recalled from an associated cue or from a fragment of itself. In other words, computers at the time stored memory like a filing cabinet. These stored memory as distributed, with every memory spread over all the connections.

In the early 1980s Kohonen’s self-organizing map brought Finnish AI research to international attention when Hopfield's 1982 network built directly on this line. Breakthroughs in parallel and distributed computing in earnest would come a decade later but this was a critical step for connectionist, learning AI in an otherwise deep winter.