Disentanglement of persistent and dynamic neural substrates of working memory
- 1Vanderbilt University
Abstract
Competing theories of working memory have alternatively emphasized the importance of persistent and dynamic substrates of this cognitive function. Here, we directly tested these theories by computationally isolating the contributions of persistent and dynamic activity to working memory. We analyzed neuronal population activity recordings in macaques doing match/non-match working memory tasks. We first extracted signatures of persistent and dynamic activity through cue-specific temporal means, temporal principal components, and correlations of activity patterns across task epochs. We then generated synthetic data constrained by these signatures, and decoded cue stimuli from these data, separately from persistent and non-persistent neurons. We found that high decoding performance arose from a nuanced interplay of persistent and dynamic activity. Specifically, we could decode stimuli at empirical or above-chance levels across all tasks only from persistent neurons, but only after constraining correlations of activity patterns across epochs. In this way, we showed distinct roles for persistent and dynamic neuronal activity that help to unify two competing theories of working memory. More broadly, we also advanced computational controlled experiments as a principled framework for adjudicating between competing models of neuronal population activity.
Related
- Working memory operations emerge from dynamic changes in neural subspace geometry
- Intrinsic cellular persistent firing sustains hippocampal spatial representations during working memory
- Neural substrates related to memory consolidation of learning multiple motor sequences during wakeful rest
- Effects of Oral L-Arginine on Everyday Persistent Fatigue, Sleep, and Post-Exercise Lactate Recovery in Community-Dwelling Working-Age Adults: A Double-Blind, Placebo-Controlled Randomized Trial
- Neural signatures of serial order access in verbal working memory
BibTeX
@misc{abbasi2026disentanglement,
title = {{Disentanglement of persistent and dynamic neural substrates of working memory}},
author = {A. Abbasi and A. Nanda and W. Dang and B. Singh and Z. Wang and C. Constantinidis and M. Rubinov},
howpublished = {bioRxiv},
year = {2026},
month = nov,
doi = {10.64898/2026.11.05.764743},
url = {https://www.biorxiv.org/content/10.64898/2026.11.05.764743v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Disentanglement of persistent and dynamic neural substrates of working memory
AU - A. Abbasi
AU - A. Nanda
AU - W. Dang
AU - B. Singh
AU - Z. Wang
AU - C. Constantinidis
AU - M. Rubinov
PY - 2026
DA - 2026/11/05
PB - bioRxiv
DO - 10.64898/2026.11.05.764743
UR - https://www.biorxiv.org/content/10.64898/2026.11.05.764743v1
AB - Competing theories of working memory have alternatively emphasized the importance of persistent and dynamic substrates of this cognitive function. Here, we directly tested these theories by computationally isolating the contributions of persistent and dynamic activity to working memory. We analyzed neuronal population activity recordings in macaques doing match/non-match working memory tasks. We first extracted signatures of persistent and dynamic activity through cue-specific temporal means, temporal principal components, and correlations of activity patterns across task epochs. We then generated synthetic data constrained by these signatures, and decoded cue stimuli from these data, separately from persistent and non-persistent neurons. We found that high decoding performance arose from a nuanced interplay of persistent and dynamic activity. Specifically, we could decode stimuli at empirical or above-chance levels across all tasks only from persistent neurons, but only after constraining correlations of activity patterns across epochs. In this way, we showed distinct roles for persistent and dynamic neuronal activity that help to unify two competing theories of working memory. More broadly, we also advanced computational controlled experiments as a principled framework for adjudicating between competing models of neuronal population activity.
ER -