Working memory representations in macaque frontal cortex
Will Reith
Project Description
Working memory (WM) allows animals to store recent information for imminent use, yet the nature of WM representations in prefrontal cortex (PFC) remains debated. Watters et al. (2025, bioRxiv) presented rhesus macaques with multiple simultaneous objects, one of which subsequently determined a rewarded saccade. They compared several models of WM representations and found that PFC activity is consistent with a gain model: objects represented by weighted activity in a shared neural pool. This contrasts with a slot model, in which objects are represented in separate subspaces - an architecture that explains WM representations in sequence-WM tasks (Xie et al., 2022, Science; El-Gaby et al., 2024, Nature; Jensen et al., 2025, bioRxiv). The discrepancy in explanatory power across disparate WM tasks may reflect differing task demands, but another possibility is that Watters et al. only tested a limited version of the slot model. Stroud et al. (2023) showed that non-sequence WM representations rotate through a series of subspaces during delay periods and recent theoretical work suggests that such rotational dynamics are both energetically and computationally efficient (Dorrell et al., 2026, bioRxiv).
We propose to reanalyse the Watters et al. (2025) dataset to test whether PFC WM representations are better described by a gain model or a dynamic slot model incorporating rotational dynamics, a comparison their original analyses did not make.
Background and References
Watters, N., Gabel, J., Tenenbaum, J. & Jazayeri, M. Working Memory of Multi-Object Scenes in Primate Frontal Cortex. 2026.01.27.702062 Preprint at https://doi.org/10.64898/2026.01.27.702062 (2026).
Xie, Y. et al. Geometry of sequence working memory in macaque prefrontal cortex. Science 375, 632–639 (2022).
Stroud, J. P., Watanabe, K., Suzuki, T., Stokes, M. G. & Lengyel, M. Optimal information loading into working memory explains dynamic coding in the prefrontal cortex. Proceedings of the National Academy of Sciences 120, e2307991120 (2023).
Dorrell, W., Latham, P. E., Behrens, T. E. J. & Whittington, J. C. R. An Efficient Computing Theory of Prefrontal Structured Working Memory Representations. 2026.02.16.706126 Preprint at https://doi.org/10.64898/2026.02.16.706126 (2026).
Jensen, K. T. et al. A mechanistic theory of planning in prefrontal cortex. 2025.09.23.677709 Preprint at https://doi.org/10.1101/2025.09.23.677709 (2025).