Learning multisensory cue integration: A computational model of crossmodal synaptic plasticity enables...

Area of Science:

• Neuroscience
• Computational Neuroscience
• Robotics

Background:

• Multisensory integration is crucial for coherent perception.
• Neural mechanisms and developmental processes of multisensory integration are not fully understood.
• Psychophysical studies suggest Bayesian principles guide multisensory cue integration based on stimulus reliability.

Purpose of the Study:

• To hypothesize and test experience-dependent crossmodal synaptic plasticity as a mechanism for developing multisensory integration.
• To develop a computational model simulating Bayesian multisensory cue integration with reliability-based weighting.

Main Methods:

• Implemented a computational model using Bayesian multisensory cue integration and reliability-based cue weighting.
• Utilized crossmodal synaptic plasticity to adapt synaptic weights based on stimulus statistics and reliability.
• Embodied the model in a simulated robotic agent for audio-visual target localization.

Main Results:

• The model successfully learned modality-specific synaptic weights proportional to the relative reliabilities of auditory and visual stimuli.
• Learned synaptic weights reflected maximum-likelihood estimation principles.
• The robotic agent effectively integrated spatial cues from multiple senses.

Conclusions:

• Experience-dependent crossmodal synaptic plasticity is a plausible mechanism for the development of multisensory integration.
• The computational model supports Bayesian principles in multisensory cue integration.
• This approach offers insights into neural plasticity and sensory processing.