Causal Inference for Cross-Modal Action Selection: A Computational Study in a Decision Making Framework.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The brain integrates multisensory information to perceive objects and events.
• Current understanding lacks a clear mechanism for how the brain compares spatiotemporal features of sensory inputs to determine their origin.
• This is crucial for distinguishing between single and multiple sources of stimuli.

Purpose of the Study:

• To propose and validate a computational model for cross-modal causal inference.
• To investigate how the brain determines if visual and auditory stimuli originate from the same source.
• To explore the role of working memory in integrating sensory information.

Main Methods:

• Developed a computational model reducing visual and auditory data to time-varying signals.
• Incorporated leaky integrators as working memory to retain sensory information.
• Utilized an evidence-based decision-making framework with saliency maps and a spatiotemporal similarity measure for causal inference.

Main Results:

• Simulations validated the model against human behavioral data in cross-modal tasks.
• The model accurately predicted behavior in novel experiments with varying stimulus features (spatial, temporal, reliability).
• The spatiotemporal similarity measure was confirmed as a viable criterion for causal inference.

Conclusions:

• The proposed model provides a viable mechanism for cross-modal causal inference and target selection in the brain.
• The spatiotemporal similarity measure is effective for inferring common or separate causes of sensory inputs.
• The framework can be extended to other cognitive tasks limited by working memory, such as complex target selection.