ELODI: Ensemble Logit Difference Inhibition for Positive-Congruent Training

arXiv:2205.06265v3 Announce Type: replace
Abstract: Negative flips are errors introduced in a classification system when a legacy model is updated. Existing methods to reduce the negative flip rate (NFR) either do so at the expense of overall accuracy by forcing a new model to imitate the old models, or use ensembles, which multiply inference cost prohibitively. We analyze the role of ensembles in reducing NFR and observe that they remove negative flips that are typically not close to the decision …

abstract accuracy arxiv classification cs.cv cs.lg difference ensemble errors negative positive rate reduce training type