Online Preplay and Offline Replay Optimize Action Selection in a Closed-Loop Navigation Model
Abstract
Hippocampal neurons not only encode the animal's current position, but also generate sequences representing potential future states and past locations. During goal-directed navigation, the represented location sweeps ahead of the animal's location (online forward replay). After such tasks, during rest, neuronal sequences occurring during sharp-wave ripples replay previously explored trajectories (offline forward replay). Following random foraging, however, neuronal sequences resemble Brownian motion, which does not correlate with behavior (diffusive replay). The mechanisms producing such diverse sequences, and their distinct effects on spatial coding and navigation, remain unclear. Here, we address these questions using a closed-loop model of navigation in a two-dimensional open-field environment based on a spiking neuronal network. In our framework, neuronal sequences are generated by a common mechanism driven by background noise and the balance of excitation and inhibition. Different replay modes arise solely from task-dependent synaptic modifications and serve different functions: online forward replay facilitates shortcut discovery, offline forward replay consolidates optimal sequences of actions, and diffusive replay helps in generalizing the action and state map and homogenizing internal connectivity. Together, these results show that a single network architecture can flexibly generate various replay types and clarify how each contributes to optimizing goal-directed navigation behavior.
Related
- Towards optimal selection schedules for online active learning
- Scaling the Wavelength to Optimize Ultrafast Laser Processing of Bandgap Materials
- A physics-informed LLM agent for uncovering typhoon- and earthquake-induced turbine failures and offline yaw-pitch countermeasures
- Modeling Transport and Mechanics in Anion Exchange Membranes to Optimize Continuous CO2 Separations
- Develop and Optimize 5DCT Imaging Simulation and Reconstruction Methods
BibTeX
@misc{ehsani2026online,
title = {{Online Preplay and Offline Replay Optimize Action Selection in a Closed-Loop Navigation Model}},
author = {Masud Ehsani and Sen Cheng},
howpublished = {arXiv},
year = {2026},
month = oct,
doi = {10.48550/arxiv.2610.26424},
eprint = {2610.26424},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.26424},
}
RIS
TY - UNPB
TI - Online Preplay and Offline Replay Optimize Action Selection in a Closed-Loop Navigation Model
AU - Masud Ehsani
AU - Sen Cheng
PY - 2026
DA - 2026/10/19
PB - arXiv
DO - 10.48550/arxiv.2610.26424
UR - https://arxiv.org/abs/2610.26424
AB - Hippocampal neurons not only encode the animal's current position, but also generate sequences representing potential future states and past locations. During goal-directed navigation, the represented location sweeps ahead of the animal's location (online forward replay). After such tasks, during rest, neuronal sequences occurring during sharp-wave ripples replay previously explored trajectories (offline forward replay). Following random foraging, however, neuronal sequences resemble Brownian motion, which does not correlate with behavior (diffusive replay). The mechanisms producing such diverse sequences, and their distinct effects on spatial coding and navigation, remain unclear. Here, we address these questions using a closed-loop model of navigation in a two-dimensional open-field environment based on a spiking neuronal network. In our framework, neuronal sequences are generated by a common mechanism driven by background noise and the balance of excitation and inhibition. Different replay modes arise solely from task-dependent synaptic modifications and serve different functions: online forward replay facilitates shortcut discovery, offline forward replay consolidates optimal sequences of actions, and diffusive replay helps in generalizing the action and state map and homogenizing internal connectivity. Together, these results show that a single network architecture can flexibly generate various replay types and clarify how each contributes to optimizing goal-directed navigation behavior.
ER -