Linear-nonlinear-time-warp-poisson models of neural activity

Area of Science:

• Computational Neuroscience
• Systems Neuroscience

Background:

• Traditional spike train models often assume only stochastic (Poisson) spiking as a source of variability.
• Neural activity may also exhibit variability due to internal cognitive processes like movement planning, which are not strictly time-locked to execution.

Purpose of the Study:

• To develop and validate a novel computational model that accounts for shared temporal variability in neural activity.
• To improve the predictive accuracy of neural firing patterns by incorporating internal timing variations.

Main Methods:

• The study integrated the standard Linear-Nonlinear-Poisson (LNP) model with Dynamic Time Warping (DTW).
• This combined approach was applied to neural recordings from the macaque premotor cortex.

Main Results:

• The integration of Dynamic Time Warping (DTW) significantly enhanced the prediction of neural activity.
• The findings indicate that temporal variability, beyond simple stochasticity, plays a crucial role in neural coding.

Conclusions:

• Temporal variability arising from internal brain processes is a significant factor in neural coding and should be explicitly modeled.
• The proposed LNP-DTW model offers a more comprehensive framework for understanding neural variability and prediction.