Real-Time Estimation of Overt Attention from Dynamic Features of the Face Using Deep Learning.

Area of Science:

• Cognitive Science
• Artificial Intelligence
• Educational Technology

Background:

• Remote learning presents challenges for monitoring student engagement due to the lack of visual feedback.
• Traditional attention assessment methods are subjective and labor-intensive.
• Inter-subject correlation (ISC) of brain activity or gaze shows promise for objective attention measurement.

Purpose of the Study:

• To develop an AI model that predicts individual student attention from facial dynamics alone.
• To eliminate the need for reference groups or manual labeling in attention assessment.
• To create a scalable, objective, and privacy-preserving tool for monitoring engagement in remote education.

Main Methods:

• Utilized Inter-Subject Correlation (ISC) of eye movements as an attention index.
• Trained a deep neural network to predict attention from a single subject's facial dynamics.
• Conducted three experiments with 83 participants.

Main Results:

• The AI model explained up to 38% of variance in known-subject data (R²=0.38) and 26-30% in new subjects (R²=0.26-0.30).
• The model captured time-resolved overt attention and correlated with post-video test scores (r=0.41-0.49).
• Eye and head movements were identified as key features driving the model's predictions.

Conclusions:

• Facial dynamics, particularly eye and head movements, can reliably predict student attention.
• The proposed AI method offers an objective, scalable, and privacy-preserving solution for real-time engagement monitoring.
• This approach enhances remote education by providing actionable insights into student focus and performance.