Balancing prior knowledge and sensory data in a predictive coding model of coherent motion detection.

This study introduces a neurobiologically inspired computational model based on the predictive coding algorithm, providing insights into coherent motion detection processes. The model is designed to reflect key principles observed in the visual system, particularly MT neurons and their surround suppression mechanisms, which play a critical role in detecting global motion. By integrating these principles, the model simulates how motion structures are decomposed into individual and shared sources, mirroring the brain's strategy for extracting coherent motion patterns. The results obtained from random dot stimuli underscore the delicate balance between sensory data and prior knowledge in motion detection. Model testing across varying noise levels reveals that, as noise increases, the model takes longer to stabilize its motion estimates, consistent with psychophysical experiments showing that response duration (e.g., reaction time or decision-making time) also increases under higher noise conditions. The model suggests that an excessive emphasis on prior knowledge prolongs the stabilization time for motion detection, whereas an optimal integration of prior expectations enhances detection accuracy and efficiency by preventing excessive disturbances due to noise. These findings contribute to potential explanations for motion detection deficiencies observed in schizophrenia.