Information Subtraction trains a generator against two mutual information estimators to represent conditional entropy H(Y|X), but the objective does not reliably remove the conditioned variable's information.
Learning bias-invariant representation by cross- sample mutual information minimization
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Information Subtraction: Learning Representations for Conditional Entropy
Information Subtraction trains a generator against two mutual information estimators to represent conditional entropy H(Y|X), but the objective does not reliably remove the conditioned variable's information.