Predictive Coding with Neural Transmission Delays: A Real-Time Temporal Alignment Hypothesis.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Hierarchical predictive coding models brain function using forward and backward connections.
• Existing models overlook neural transmission delays, causing prediction misalignment with time-varying stimuli.
• This temporal lag poses challenges for real-time sensory processing, particularly for dynamic events like visual motion.

Purpose of the Study:

• To extend predictive coding models by incorporating neural transmission time.
• To develop a model that minimizes prediction errors by realigning backward predictions.
• To investigate the implications of temporal alignment for neural representations and perception.

Main Methods:

• Developed an extended hierarchical predictive coding model with forward and backward extrapolation mechanisms.
• Simulated the model using visual motion as a test case.
• Evaluated model plausibility, consistency with existing neuroscience evidence, and predictive power for perceptual phenomena.

Main Results:

• The extended model successfully realigns backward predictions to minimize prediction error in real time.
• Neural representations across hierarchical levels are synchronized.
• The model explains several known motion-position illusions and offers a solution to the temporal binding problem.

Conclusions:

• The proposed model offers a neurally plausible account of real-time sensory processing by addressing temporal delays.
• Extrapolation mechanisms are crucial for aligning predictions with sensory input throughout the visual hierarchy.
• This work advances our understanding of cortical computation and perceptual phenomena like visual motion illusions.