A biologically inspired neural model for visual and proprioceptive integration including sensory training

Area of Science:

• Neuroscience
• Computational Neuroscience
• Human Perception

Background:

• Multisensory integration models traditionally use Bayesian inference for single or dual causal sources.
• Previous research explored proprioceptive training's impact on multisensory perception.
• The role of training in multisensory integration remains under-explored.

Purpose of the Study:

• To present a new recurrent neural model for integrating visual and proprioceptive information using population coding.
• To simulate and compare the effects of visual versus proprioceptive training on multisensory integration.
• To investigate the sensory training process in human multisensory perception.

Main Methods:

• Developed a recurrent neural network model based on population coding to mimic human brain multisensory integration.
• Conducted experiments with human subjects performing circular hand movements and position estimation.
• Assigned subjects to either visual training or proprioceptive training groups (eight subjects each).

Main Results:

• The neural model's simulation results align with causal Bayesian inference.
• Visual training exhibited a significantly higher learning rate compared to proprioceptive training.
• Both visual and proprioceptive errors decreased with training; this reduction was statistically significant for proprioception but not for vision.

Conclusions:

• The developed neural model effectively simulates multisensory integration and training processes.
• Visual training offers faster learning and more significant error reduction in multisensory tasks than proprioceptive training.
• Experimental findings support the neural model's simulation outcomes regarding training effects on perception.