1 |
Face structure, color, and texture, and their weighted combination are reliable predictors of facial affect; each metric varies by gender in an expected manner |
Validates model; extends previous work showing gender differences in facial structure to texture and color as well |
2 |
All three metrics correlate with, and in some cases, mediate the relationship between face gender and human impressions
|
Provides correlational evidence that the metrics used by machine learning to predict emotions relates to human impressions in an expected manner |
3 |
Algorithmically-derived impressions of dominance and affiliation are related to human impressions of dominance and trustworthiness |
Demonstrates that higher-order impressions can be derived from machine learning output trained on emotions |
4 |
Algorithmically-derived impressions can be used to reverse-engineer important structure, color, and texture features in neutral faces |
Experimentally demonstrates that metrics machine learning models use to predict emotions are also used by humans to form impressions
|