Uncertainty modeling in affective computing

Asma Ghandeharioun, Brendan Jou, Brian Eoff, Florian Schroff

Abstract: This disclosure describes techniques that capture the uncertainty in machine-vision based affect (emotion) perception. The techniques are capable of predicting aleatoric, epistemic, and annotation uncertainty. Measures of uncertainty are important to safety-critical and subjective assessment tasks such as those found in the perception of affective expressions.