A benchmark of XAI methods for VLMs models using faithfulness metrics (including a novel faithfulness metric to measure cross-modal reasoning in VLMs).

Membership Inference Attacks on Tabular Foundations Models

Counterfactual explanations for multivariate time series classifiers.

A unified benchmark dataset combining HumanEval, MBPP, and robustness-focused variants from multiple papers to evaluate how well LLMs handle imperfect programming task descriptions, including ambiguous, incomplete, contradictory... The dataset supports research on code generation robustness and reliability under real world task conditions.

A reasoning benchmark for LLMs, comprising graph and tabular tasks.