Multiple mechanisms of rhythm switching in recurrent neural networks with adaptive time constants
Abstract
Although recurrent neural networks (RNNs) trained on cognitive tasks have become a widely used framework for studying neural computation, the internal mechanisms by which RNNs switch between rhythms across multiple frequency bands, and how these mechanisms relate to neuronal time constants, have not been systematically analyzed. We trained leaky integrator RNNs with neuron-specific learnable time constants on a four-band (theta, alpha, beta, gamma) rhythm-switching task and analyzed 20 independently trained networks. Whereas low-frequency rhythms were produced by distributed participation of many neurons, high-frequency rhythms were dominated by a small subpopulation of short-time-constant neurons, and the negative correlation between time constant and matched-mode amplitude strengthened monotonically with frequency. Rhythm switching was supported by multiple coexisting mechanisms: turnover of the active subpopulation, network-wide baseline shifts that reposition the operating point near distinct unstable fixed points, and inter-neuronal phase reorganization that selectively cancels or supports band components in the population output. Analysis of the learned recurrent connectivity linked these mechanisms to network structure: the short-time-constant neurons formed a more strongly interconnected module, and the baseline shift retuned the oscillation frequency through gain modulation of the recurrent interactions. The mechanism deployed for each mode pair varied across training runs, exposing a degeneracy of learned solutions. These findings parallel the coexistence of rhythm-specific and multi-rhythm interneurons reported in biological circuits and provide a candidate framework for interpreting frequency-band-specific functional differentiation in neural systems.
Versions
Related
- Shaping Slow Manifolds in Equivariant Recurrent Neural Networks
- Reusable modular architecture enables flexible cognitive operations in the mouse brain and artificial recurrent networks
- Ameliorating effects and mechanisms of therapeutic electrical stimulation on the recurrent laryngeal nerve injury model in rats
- Language usage shapes neural mechanisms of code-switching
- Heterogeneous plasticity rules generate diverse structures in recurrent networks
BibTeX
@misc{yamaguti2026multiple,
title = {{Multiple mechanisms of rhythm switching in recurrent neural networks with adaptive time constants}},
author = {Yutaka Yamaguti and Shota Nakamura},
howpublished = {arXiv},
year = {2026},
month = aug,
doi = {10.48550/arxiv.2605.14388},
eprint = {2605.14388},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2605.14388},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Multiple mechanisms of rhythm switching in recurrent neural networks with adaptive time constants
AU - Yutaka Yamaguti
AU - Shota Nakamura
PY - 2026
DA - 2026/08/12
PB - arXiv
DO - 10.48550/arxiv.2605.14388
UR - https://arxiv.org/abs/2605.14388
AB - Although recurrent neural networks (RNNs) trained on cognitive tasks have become a widely used framework for studying neural computation, the internal mechanisms by which RNNs switch between rhythms across multiple frequency bands, and how these mechanisms relate to neuronal time constants, have not been systematically analyzed. We trained leaky integrator RNNs with neuron-specific learnable time constants on a four-band (theta, alpha, beta, gamma) rhythm-switching task and analyzed 20 independently trained networks. Whereas low-frequency rhythms were produced by distributed participation of many neurons, high-frequency rhythms were dominated by a small subpopulation of short-time-constant neurons, and the negative correlation between time constant and matched-mode amplitude strengthened monotonically with frequency. Rhythm switching was supported by multiple coexisting mechanisms: turnover of the active subpopulation, network-wide baseline shifts that reposition the operating point near distinct unstable fixed points, and inter-neuronal phase reorganization that selectively cancels or supports band components in the population output. Analysis of the learned recurrent connectivity linked these mechanisms to network structure: the short-time-constant neurons formed a more strongly interconnected module, and the baseline shift retuned the oscillation frequency through gain modulation of the recurrent interactions. The mechanism deployed for each mode pair varied across training runs, exposing a degeneracy of learned solutions. These findings parallel the coexistence of rhythm-specific and multi-rhythm interneurons reported in biological circuits and provide a candidate framework for interpreting frequency-band-specific functional differentiation in neural systems.
ER -