A Dynamic Neural Field Model of Multimodal Merging: Application to the Ventriloquist Effect

Area of Science:

• Computational neuroscience
• Sensory integration
• Psychophysics

Background:

• Multimodal merging integrates imprecise sensory information (e.g., sight, hearing) for localization.
• Bayesian models traditionally describe merging decisions based on aggregated behavioral data.
• Existing models lack trial-by-trial dynamic simulation capabilities.

Purpose of the Study:

• Introduce a novel computational model using dynamic neural fields for multimodal merging.
• Simulate trial-by-trial localization decisions adapting to audio-visual discrepancies.
• Compare model performance and dynamics against established Bayesian models.

Main Methods:

• Developed a dynamic neural field model simulating mesoscopic neural processes.
• Fit the model to human psychophysical data, specifically the ventriloquist effect.
• Investigated the influence of retinotopic projection on the superior colliculus.

Main Results:

• The dynamic neural field model simulates localization decisions trial-by-trial.
• Model performance is comparable to Bayesian models on average.
• Qualitative analysis revealed insights into decision dynamics and noise-induced perceptual variations.

Conclusions:

• Dynamic neural fields offer a viable alternative to Bayesian models for simulating multimodal merging dynamics.
• The model's parameter space allows flexibility for fitting specific criteria or enhancing performance.
• This approach has potential applications in simulating behavioral dynamics and visually guided saccades.