Learning to Control Rapidly Changing Synaptic Connections: An Alternative Type of Memory in Sequence Processing Artificial Neural Networks
This work addresses memory mechanisms in sequence processing for AI and neuroscience, but it is incremental as it revives and reframes existing FWP concepts.
The paper tackles the problem of short-term memory in sequence-processing neural networks by proposing Fast Weight Programmers (FWPs) as an alternative to standard recurrent neural networks, achieving competitive performance across various tasks. It presents FWPs in the context of artificial neural networks as an abstraction of biological neural networks, a perspective not emphasized in prior work.
Short-term memory in standard, general-purpose, sequence-processing recurrent neural networks (RNNs) is stored as activations of nodes or "neurons." Generalising feedforward NNs to such RNNs is mathematically straightforward and natural, and even historical: already in 1943, McCulloch and Pitts proposed this as a surrogate to "synaptic modifications" (in effect, generalising the Lenz-Ising model, the first non-sequence processing RNN architecture of the 1920s). A lesser known alternative approach to storing short-term memory in "synaptic connections" -- by parameterising and controlling the dynamics of a context-sensitive time-varying weight matrix through another NN -- yields another "natural" type of short-term memory in sequence processing NNs: the Fast Weight Programmers (FWPs) of the early 1990s. FWPs have seen a recent revival as generic sequence processors, achieving competitive performance across various tasks. They are formally closely related to the now popular Transformers. Here we present them in the context of artificial NNs as an abstraction of biological NNs -- a perspective that has not been stressed enough in previous FWP work. We first review aspects of FWPs for pedagogical purposes, then discuss connections to related works motivated by insights from neuroscience.