Slow manifolds in recurrent networks encode working memory efficiently and robustly

TL;DR

This study reveals that recurrent networks encode working memory along slow stable manifolds, offering a more efficient and noise-robust mechanism compared to traditional stable attractor models.

Contribution

It demonstrates that slow manifold encoding is a prevalent and advantageous dynamical mechanism for working memory in recurrent networks, supported by extensive modeling and analysis.

Findings

Memory encoding along slow stable manifolds leads to phasic activation.

Networks naturally forget stimuli over time, yet remain robust to noise.

Slow manifold mechanisms are more efficient than stable attractor models.

Abstract

Working memory is a cognitive function involving the storage and manipulation of latent information over brief intervals of time, thus making it crucial for context-dependent computation. Here, we use a top-down modeling approach to examine network-level mechanisms of working memory, an enigmatic issue and central topic of study in neuroscience and machine intelligence. We train thousands of recurrent neural networks on a working memory task and then perform dynamical systems analysis on the ensuing optimized networks, wherein we find that four distinct dynamical mechanisms can emerge. In particular, we show the prevalence of a mechanism in which memories are encoded along slow stable manifolds in the network state space, leading to a phasic neuronal activation profile during memory periods. In contrast to mechanisms in which memories are directly encoded at stable attractors, these networks naturally forget stimuli over time. Despite this seeming functional disadvantage, they are more efficient in terms of how they leverage their attractor landscape and paradoxically, are considerably more robust to noise. Our results provide new dynamical hypotheses regarding how working memory function is encoded in both natural and artificial neural networks.