The paper investigates the phenomenon of explanation multiplicity in SHAP, showing that SHAP explanations can vary significantly across repeated runs even when the model, input, and prediction remain unchanged. The authors introduce a dual‑seed evaluation protocol, a hierarchy of metrics, and null models to quantify this variability, finding that multiplicity is widespread across datasets, models, and sampling strategies. They also demonstrate that common metrics like L2 distance may underestimate instability, while rank‑based metrics reveal substantial changes in top‑ranked features, and that improved sampling methods do not fully eliminate run‑to‑run variation.
arXiv:2608. 11508v1 Announce Type: new Abstract: Machine learning pipelines commonly flatten relational data into single-table representations, discarding structural constraints.
The paper proposes a new evaluation test for explanation methods: if an explanation accurately captures how a model uses its features, one should be able to reconstruct the model’s predictions from it. The authors convert explanations into predictors by summing feature effects and assess how well these predictors reproduce the model on unseen data, without any fitting. They apply this test to partial dependence plots, accumulated local effects, SHAP, and LIME across multiple datasets and model families, showing that the best method depends on feature dependence and that some existing quality metrics can favor flawed explanations.
arXiv:2512. 22240v5 Announce Type: replace-cross Abstract: Machine learning models are primarily judged by predictive performance, especially in applied genomics, where explanations are read as biological findings.
The paper introduces a straightforward evaluation method for explanation techniques: by converting each explanation into a predictor that sums the feature effects, the authors assess how accurately this predictor reproduces the original model’s predictions on unseen data. This approach applies to any explanation expressible as a function of features and is demonstrated on PDP, ALE, SHAP, and LIME. The authors theoretically show that summing partial dependence curves yields the optimal additive summary when features are independent, but this property fails with dependent features, and empirical results across diverse datasets confirm that the best-performing method depends on feature dependence.
arXiv:2603. 14894v3 Announce Type: replace-cross Abstract: Trust and ethical concerns due to the widespread deployment of opaque machine learning (ML) models motivating the need for reliable model explanations.
arXiv:2605.30018v3 Announce Type: replace-cross Abstract: Large language models (LLMs) frequently achieve impressive scores on standardized benchmarks, yet accuracy alone offers a limited view of the...
The paper introduces First-Order In-Context Shapley, an inference‑only method that evaluates data utility for large language model alignment by computing the first‑order term of an in‑context cooperative game over semantic k‑NN neighborhoods in O(k) forward passes. It projects log‑likelihood shifts onto a directed influence graph, yielding three signals—Exemplifying Advantage, Exemplified Advantage, and One‑of‑a‑Kind Score—to identify high‑utility demonstrations and rare prompts. The approach reduces compute by 19× compared to Monte‑Carlo Shapley, filters 1–5% of records, and improves error detection and reward accuracy in DPO training and benchmark auditing. whyItMatters":"By enabling scalable, inference‑only data auditing, the method addresses label noise and safety inversions that hinder post‑training alignment of large language models, leading to measurable improvements in model reliability and reward accuracy."
arXiv:2605. 27618v2 Announce Type: replace Abstract: Despite the wide use of explainability techniques to attempt to understand the behavior of Artificial Intelligence (AI), the generated explanations may not always be reliable.
arXiv:2607. 03675v1 Announce Type: new Abstract: Shapley values are widely used to attribute value to training data based on their marginal contribution to performance on a validation set.
The paper investigates how the definition of influence—specifically the behavior being attributed, the intervention on training data, and the counterfactual training process—affects rankings produced by influence estimators. It formalizes influence as a counterfactual estimand, distinguishes specification mismatch from approximation error, and categorizes existing estimators by their implied specifications. Experiments demonstrate that different specifications can lead to markedly different rankings, and that careful specification choice improves attribution quality in tasks such as noisy label detection and large‑language‑model attribution.
arXiv:2607. 22045v1 Announce Type: new Abstract: Counterfactual explanations are a prominent approach in explainable artificial intelligence (xAI), providing actionable guidance on what input changes would alter a model's prediction to a desired outcome.
AMLworld-Compact introduces an importance‑weighted downsampling protocol for evaluating large language models on AMLworld’s test splits, retaining all illicit transactions while sampling benign ones from supervised‑score strata. The HT‑Coreset method reduces evaluation targets by 271× on HI‑Small and 611× on LI‑Small, providing unbiased confusion‑count estimates with exact recall and ratio estimates for precision and F1. Experiments with boosted‑tree strata and Graph Feature Preprocessor features show lower F1 RMSE compared to uniform sampling, and audit tools reveal that lexical checks do not reliably indicate correctness of model outputs.