Subjective Risk Decomposition: A New View for Uncertainty Quantification
The paper treats uncertainty measures as outputs of modeling choices, not standalone objects.
Alamri, Caprio, and Brown derive epistemic and aleatoric uncertainty by decomposing a subjective risk built from a strictly proper loss. Reverse cross-entropy is used as a key example, recovering classic information-theoretic terms. They argue the same framework also recovers many uncertainty measures already proposed in the literature. The paper extends the setup into learning theory with subjective-risk versions of excess risk, approximation error, and estimation error. ArXiv · AI/CL/LG's note
Alamri, Caprio, and Brown derive epistemic and aleatoric uncertainty by decomposing a subjective risk built from a strictly proper loss. Reverse cross-entropy is used as a key example, recovering classic information-theoretic terms. They argue the same framework also recovers many uncertainty measures already proposed in the literature. The paper extends the setup into learning theory with subjective-risk versions of excess risk, approximation error, and estimation error. ArXiv · AI/CL/LG's note
score 4