In this paper, we introduce the counterfactual score as a new evaluation metric for black-box abstaining classifiers.

The reasons for this are clear.

Multimodal (MM) models aim to extract the synergistic predictive potential of multiple data types to create a shared feature space with aligned semantic meaning across inputs of drastically varying sizes (i.e.

LM significantly improves the effectiveness and efficiency of LLMs.

In particular, it is vital to minimize a model's memorization of sensitive