Causality in Cognitive Neuroscience: Concepts, Challenges, and Distributional Robustness

Area of Science:

• Cognitive Neuroscience
• Causal Inference
• Machine Learning

Background:

• Probabilistic models capture variable dependencies.
• Causal models predict effects of interventions on cognitive functions and neuronal activity.
• Inferring causality in cognitive neuroscience is complex.

Purpose of the Study:

• Introduce causality concepts in cognitive neuroscience.
• Review methods for inferring causal relationships from data.
• Propose distributional robustness as a guiding principle.

Main Methods:

• Review of existing causal inference methods.
• Discussion of challenges: scarcity of interventional data and variable selection.
• Argument for distributional robustness (invariance) as a solution.

Main Results:

• Causal models must generalize across environments when mechanisms are invariant.
• Lack of generalization indicates incorrect causes or variable granularity.
• Generalizability assessment aids variable definition and compensates for limited interventional data.

Conclusions:

• Distributional robustness is key for causal inference in cognitive neuroscience.
• Assessing model generalizability is crucial for identifying correct causal mechanisms and variables.
• This principle helps overcome data limitations in the field.