Exploring the suitability of piecewise-linear dynamical system models for cognitive neural dynamics DOI Creative Commons
Jiemin Wu, Boateng Asamoah, Zhaodan Kong

et al.

Frontiers in Neuroscience, Journal Year: 2025, Volume and Issue: 19

Published: May 12, 2025

Dynamical system models have proven useful for decoding the current brain state from neural activity. So far, neuroscience has largely relied on either linear or non-linear based artificial networks (ANNs). Piecewise approximations of dynamics in other technical applications. Moreover, such explicit provide a clear advantage over ANN-based when dynamical is not only supposed to be observed, but also controlled, particular controller with guarantees needed. Here we explore whether piecewise-linear (recurrent Switching Linear System rSLDS models) could modeling dynamics, context cognitive tasks. These that they can estimated continuous observations like field potentials smoothed firing rates, sparser single-unit spiking data. We first generate data computational model perceptual decision-making and demonstrate successfully recovered these observations. then outperforms terms predicting future states associated Finally, apply our approach publicly available dataset recorded monkeys performing decisions. Much surprise, did significant data, although were different trial epochs showed qualitatively dynamics. In summary, present prove situations, where needs controlled closed-loop fashion, example, new neuromodulation applications treating deficits. Future work will show under what conditions are sufficiently warrant use one.

Language: Английский

Exploring the suitability of piecewise-linear dynamical system models for cognitive neural dynamics DOI Creative Commons
Jiemin Wu, Boateng Asamoah, Zhaodan Kong

et al.

Frontiers in Neuroscience, Journal Year: 2025, Volume and Issue: 19

Published: May 12, 2025

Dynamical system models have proven useful for decoding the current brain state from neural activity. So far, neuroscience has largely relied on either linear or non-linear based artificial networks (ANNs). Piecewise approximations of dynamics in other technical applications. Moreover, such explicit provide a clear advantage over ANN-based when dynamical is not only supposed to be observed, but also controlled, particular controller with guarantees needed. Here we explore whether piecewise-linear (recurrent Switching Linear System rSLDS models) could modeling dynamics, context cognitive tasks. These that they can estimated continuous observations like field potentials smoothed firing rates, sparser single-unit spiking data. We first generate data computational model perceptual decision-making and demonstrate successfully recovered these observations. then outperforms terms predicting future states associated Finally, apply our approach publicly available dataset recorded monkeys performing decisions. Much surprise, did significant data, although were different trial epochs showed qualitatively dynamics. In summary, present prove situations, where needs controlled closed-loop fashion, example, new neuromodulation applications treating deficits. Future work will show under what conditions are sufficiently warrant use one.

Language: Английский

Citations

0