Abstract
Abstract
Human working memory enables us to temporarily store and flexibly utilize information. It is thought that there are two distinct neural mechanisms by which information is retained in working memory: persistent firing of working memory-relevant neurons
1
, or short-term synaptic plasticity in the absence of ongoing neuronal firing
2
. However, both these mechanisms can lead to strong neural interference when working memory contents are updated with novel information. As an alternative, here we propose that information can be flexibly held in working memory in the form of traveling waves - spatiotemporal patterns of neural activity that propagate across brain tissue. In an artificial neural network, we simulated wave propagation by passing transient neural firing patterns representative of single-item information across different network layers. In comparison to sustained neural firing, we demonstrate that these transient traveling waves result in minimal synaptic plasticity, thereby facilitating working memory updating without significant neural interference. We then identified such traveling waves in intracranial recordings from the human temporal lobe. Traveling waves precisely when information was being encoded and persisted uniformly throughout maintenance and recall, aligning with the activation pattern seen in the artificial neural network. In trials where participants committed memory errors, no such continuous traveling wave activity in the temporal lobe was observed. Our data from artificial neural networks and human intracranial brain recordings converge to indicate that neural traveling waves support information processing in working memory while minimizing interference during updating. This here reported mechanism, together with the classical working memory models, serves as a mutually complementary, multi-mechanistic framework for working memory storage.