Slow manifolds in recurrent networks encode working memory efficiently and robustly
This research provides new dynamical hypotheses for how working memory is encoded in both natural and artificial neural networks, addressing an enigmatic issue in neuroscience and machine intelligence.
This paper investigates working memory mechanisms in recurrent neural networks (RNNs) trained on a working memory task. They found that memories encoded along slow stable manifolds lead to efficient and robust working memory, despite natural forgetting over time. These networks are more efficient in leveraging their attractor landscape and are considerably more robust to noise compared to networks encoding memories at stable attractors.
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.