A biologically inspired neurocomputational model for audiovisual integration and causal inference

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The brain effectively extracts information from noisy sensory data and performs causal inference.
• Understanding the neural architecture for multisensory integration and causal inference remains a challenge.
• Existing research suggests the nervous system employs statistically optimal, probabilistic strategies.

Purpose of the Study:

• To develop a physiologically plausible mathematical model of neural mechanisms for multisensory perception and causal inference.
• To analyze how cross-modal interactions are implemented in neural networks.
• To investigate the brain's architecture for processing noisy sensory environments.

Main Methods:

• A three-layered neural network model was designed with separate auditory and visual input layers.
• Reciprocal excitatory connections between input layers and feedforward connections to a third layer were implemented.
• The model's outputs were compared against existing behavioral data on multisensory perception.

Main Results:

• The model successfully replicated phenomena like the ventriloquism illusion and sensory bias.
• It explained the percept of unity based on auditory-visual spatial distance.
• Simulation results aligned with behavioral data, indicating probability matching as a key strategy in auditory-visual localization.

Conclusions:

• The developed neural network model provides a framework for understanding multisensory perception and causal inference.
• The model's ability to account for specific illusions and biases supports its physiological plausibility.
• The study offers testable predictions for future behavioral experiments on auditory-visual integration.