Neural networks learn highly selective representations in order to overcome the superposition catastrophe.

28 citations recorded by the source corpus.

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

On the Binding Problem in Artificial Neural Networks (2020, background): “In fact, even for plain RNNs it was found that when they are trained to remember multiple objects internally, they resort to a localist representation (Bowers et al., 2014).”
Grandmother cells and localist representations: a review of current thinking (2017, background): “As reviewed above, Bowers et al. (2014) found that recurrent PDPmodels learn localist representations when trained to co-activate multiple items at the same time in STM, but learned distributed representations when trained on the same items one-at-time.”