Decoding binary decisions under differential target probabilities from pupil dilation: A random forest approach.

Area of Science:

• Cognitive Neuroscience
• Human-Computer Interaction
• Machine Learning

Background:

• Pupil dilation is linked to cognitive processes, including visual attention and intention.
• Previous studies on decoding intentions from pupillary dynamics lacked strict experimental control for confounds.

Purpose of the Study:

• To investigate the feasibility of decoding covert intentions from pupillary dynamics using a machine learning approach.
• To rigorously control for potential confounds in pupillary response measurements.

Main Methods:

• Analysis of pupillary data from 69 participants across 19,417 trials with varying target/distractor probabilities.
• Application of a machine learning model to decode binary intentions (target vs. distractor).
• Comparison of feature importances to identify informative aspects of the pupillary signal.

Main Results:

• Pupil dilation reliably indicates observer intentions, confirming previous findings.
• Machine learning achieved up to 76% Area Under the Curve (AUC) for decoding intentions when targets were rarer than distractors.
• The first derivative of pupil size changes was the most significant predictor, enabling decoding within approximately 800ms.

Conclusions:

• Decoding intentions from pupil dilation is feasible with high accuracy under controlled conditions.
• Pupillary dynamics, particularly the rate of change, offer a promising biomarker for covert intentions.
• Findings support potential applications in visual search, gaze-based interaction, and human-robot interaction.