Assessing Cognitive Performance Using Physiological and Facial Features: Generalizing Across ...
summary
123 participants across four very different tasks (Pac‑Man, adaptive assessment, code debugging, gaze‑controlled game) produce facial action units and VGG16 frame features plus physiological time‑series summaries (value and spectral histograms, ARMA‑style models). An ensemble pipeline with LASSO, random forest and PCA uses cross‑training and ANOSIM to compute a feature generalizability index (R and p). Results show training on multiple datasets yields generalizable predictors; variance features matter most and a reduced generalizable feature set predicts held‑out gaze performance with under 10% error.
AI-written summary · about
chapters (8)
- 0:00Intro
- 0:19What?
- 2:04How?
- 2:44Datasets
- 3:32Feature Extraction
- 4:18Pipelines
- 5:01Results: Generalizable features
- 6:10What does all this mean?
related talks
23mPrediction of music pairwise preferences from facial expressions
49mAffective Computing – Bringing Humans and Machines Closer Through Emotions
21mFramework for Electroencephalography-based Evaluation of User Experience
31mIndividual Gaps in Prediction Generalization
20mBuilding a Personalized, Auto-Calibrating Eye Tracker from User Interactions
03mIdentifying Children Metacognitive Monitoring Performance Through Facial Expressions
39mBuilding Highly Scalable Facial Recognition Pipelines
19mWhat Can Be Predicted from Six Seconds of Driver Glances?