English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Working memory is a cognitive function involving the storage and manipulationof latent information over brief intervals of time, thus making it crucial forcontext-dependent computation. Here, we use a top-down modeling approach toexamine network-level mechanisms of working memory, an enigmatic issue andcentral topic of study in neuroscience and machine intelligence. We trainthousands of recurrent neural networks on a working memory task and thenperform dynamical systems analysis on the ensuing optimized networks, whereinwe find that four distinct dynamical mechanisms can emerge. In particular, weshow the prevalence of a mechanism in which memories are encoded along slowstable manifolds in the network state space, leading to a phasic neuronalactivation profile during memory periods. In contrast to mechanisms in whichmemories are directly encoded at stable attractors, these networks naturallyforget stimuli over time. Despite this seeming functional disadvantage, theyare more efficient in terms of how they leverage their attractor landscape andparadoxically, are considerably more robust to noise. Our results provide newdynamical hypotheses regarding how working memory function is encoded in bothnatural and artificial neural networks.

Slow manifolds in recurrent networks encode working memory efficiently and robustly