Quantifying Cognitive Load Via Neural Oscillatory Signatures During Predictive Coding Mismatch Error Signals in Simulated Avian Navigation

Introduction

The brain operates not as a passive receiver of sensory input, but as an active predictive engine, constantly generating hypotheses about upcoming sensory data. Navigational tasks, particularly those requiring course correction amidst sensory noise, place immense demands on this predictive coding mechanism [1]. When the actual sensory input deviates significantly from the prediction, a ‘mismatch error’ signal must be generated and processed. In biological systems, this mismatch error is hypothesized to carry the critical information regarding prediction success or failure, demanding substantial cognitive resources proportionate to the magnitude of the prediction failure.

Our research focuses on quantifying the neural signature accompanying this error signal generation during simulated avian migratory tasks. Avian navigation provides a compelling biological model due to its remarkable robustness and adaptability to noisy, dynamic environments. We operationalize cognitive load not merely as the passage of time, but as the energetic cost required to resolve prediction uncertainty. This cost is hypothesized to be directly encoded in the oscillatory dynamics between lower-frequency theta rhythms (responsible for context maintenance) and higher-frequency gamma rhythms (responsible for detailed feature binding) [2].

Theoretical Framework: Predictive Coding and Oscillatory Dynamics

We adopt a framework rooted in Predictive Coding theory, where sensory input ($\mathbf{s}$) is compared against the brain’s internal generative model ($\hat{\mathbf{s}}$), producing an error term ($\mathbf{e}$): $\mathbf{e} = \mathbf{s} - \hat{\mathbf{s}}$. Successful prediction minimizes $\mathbf{e}$, while large $|\mathbf{e}|$ drives updates to the generative model. In our neural simulation, these error signals are mapped onto local field potential fluctuations.

We hypothesize that the efficiency of error resolution—and thus the cognitive load—is modulated by the phase-amplitude coupling ($\text{PAC}$) between $\theta$ and $\gamma$ bands. Specifically, high predictive confidence corresponds to tight, phase-locked information transfer; cognitive saturation, however, manifests as a breakdown in this temporal coherence, suggesting the system is attempting to process an error signal that exceeds its immediate update capacity. This saturation point ($\text{CL}_{\text{sat}}$) can be modeled by examining the power spectral density ratio of the observed activity.

$$\text{CL} \propto -\frac{d\text{PAC}(\theta, \gamma)}{dt} \quad \text{when } |\mathbf{e}| > \epsilon_{\text{thresh}}$$

Where $\epsilon_{\text{thresh}}$ is the minimum detectable mismatch error threshold, and the derivative indicates the rate of coherence decay associated with sustained, high-magnitude error processing [3].

Computational Methodology and Simulation Parameters

Our simulation utilizes a spatially distributed neural network model comprising populations analogous to the avian hippocampus and related forebrain structures, simulated in a high-performance computing cluster. The network integrates spiking activity models with continuous firing rate approximations necessary for phase synchronization analysis. The task paradigm involves subjects navigating a simulated environment while receiving intermittent, low-bandwidth positional updates corrupted by simulated environmental noise. The primary variable manipulated across trials was the Magnitude of Prediction Error ($\text{MPE}$), ranging from near-zero (perfect prediction) to extreme (catastrophic navigational drift).

Table 1. Simulated Cognitive Load Markers

We tracked the $\text{PLI}_{\theta\gamma}$ as the proxy for successful error encoding, which quantifies the correlation between the phase of the $\theta$ oscillation and the peak of the $\gamma$ burst activity. This metric successfully differentiated between recoverable prediction errors and those overwhelming the system’s capacity to generate a coherent update signal [2].

Results

The simulation data strongly supports the hypothesis linking $\text{PLI}_{\theta\gamma}$ decay to increasing cognitive load during high-error correction phases. As the injected $\text{MPE}$ approached the critical threshold ($>0.8$), the $\text{PLI}_{\theta\gamma}$ exhibited a non-linear, rapid decline. This decline was disproportionately steeper than the linear decay observed in naive models lacking the dedicated error-scaling mechanism.

Furthermore, analysis of the underlying firing patterns revealed that subjects operating near $\text{CL}_{\text{sat}}$ exhibited increased firing variability across principal neuron populations, suggesting a resource bottleneck shift from signal processing to noise management. The relationship between the sustained mismatch signal strength and the required firing rate scaling ($u$) can be crudely modeled as:

$$\nu_{\text{required}} = \nu_{\text{baseline}} \cdot \left(1 + k \cdot \text{MPE}\right)$$

Where $k$ is a scaling constant derived from the network’s intrinsic plasticity limits [4].

Discussion and Conclusion

The identification of $\text{PLI}_{\theta\gamma}$ decay as a reliable metric for cognitive load during predictive error processing offers a powerful computational analogy for studying higher-order cognition. The system does not linearly equate error magnitude to processing cost; rather, the cost scales with the rate at which the system must sustain mismatch detection against its intrinsic noise floor. The observed breakdown suggests that sustained, high-magnitude prediction errors force the network into a sustained, high-energy maintenance mode, characteristic of cognitive saturation rather than transient error correction.

Future work will involve testing the recovery dynamics following saturation. If the environment stabilizes, we predict a measurable reversal of the $\text{PLI}_{\theta\gamma}$ decay curve, corresponding to the successful reprocessing and integration of the catastrophic error into a revised generative model, effectively resetting the cognitive load counter [1].