Log-Odds
The logarithmic ratio of probabilities between two classes, defined as log[P(Y=1|X)/P(Y=0|X)]. It serves as a natural measure in discriminative classification, with its sign directly determining the classification outcome. Log-odds occupies a central role in discriminative models such as logistic regression.
Timeline (last 90 days)
The paper proposes a discriminative reformulation that uses log-odds as the basis for supervised representations, replacing the traditional conditional log-likelihood.
The paper proves that the distance corresponding to the representation induced by log-odds is exactly equal to the ℓ1 distance between analytical Shapley value vectors.
All Facts (2)
The paper proposes a discriminative reformulation that uses log-odds as the basis for supervised representations, replacing the traditional conditional log-likelihood.
50%UnverifiedThe paper proves that the distance corresponding to the representation induced by log-odds is exactly equal to the ℓ1 distance between analytical Shapley value vectors.
50%