Different computations over the same inputs produce selective behavior in algorithmic brain networks

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Neuroscience

Background:

• Understanding how human brain networks compute sensory inputs to generate behavior is a fundamental challenge in neuroimaging.
• Dynamic algorithms underlying neural computation remain poorly understood, especially concerning the 'how' of information processing.

Purpose of the Study:

• To investigate the spatiotemporal dynamics of neural computations underlying distinct logical functions (XOR, OR, AND) using magnetoencephalography (MEG).
• To identify the specific stages and brain regions involved in processing identical inputs differently to produce varied behavioral outputs.

Main Methods:

• Recorded magnetoencephalography (MEG) data from participants performing XOR, OR, and AND behavioral tasks (N=10 per task, N-of-1 replications).
• Applied source localization to MEG activity to track neural signal progression through computational stages.
• Analyzed spatiotemporal patterns to differentiate task-specific computations.

Main Results:

• Identified four distinct computational stages in source-localized MEG activity within individual participants.
• Observed initial input representation in occipital cortex, followed by linear combination in midline occipital cortex and right fusiform gyrus.
• Found nonlinear, task-dependent integration in temporal-parietal cortex and response representation in postcentral gyrus.
• Demonstrated that early computational stages were similar across tasks, while later stages were task-specific.

Conclusions:

• Dynamic network algorithms in the human brain process identical inputs differently based on the required computation.
• Revealed the specific spatiotemporal dynamics of neural computations, detailing where, when, and how different behaviors emerge from sensory inputs.