arXiv:2607. 23804v1 Announce Type: cross Abstract: Context attribution methods for large language models (LLMs) identify which input context contributes to the model response.
arXiv:2608. 10537v1 Announce Type: new Abstract: Sparse autoencoders (SAEs) have helped uncover mechanistic explanations for LLM behaviours such as reasoning, jailbreaking etc.
The paper introduces Ra‑NEM, a method that optimizes the area under insertion and deletion curves to improve the faithfulness of feature importance attributions in machine learning models. By deriving an objective function from faithfulness and approximating its gradient, the authors link insertion curves to top‑k feature selection, enabling a loss function that measures attribution quality. Ra‑NEM, integrated with the neural explanation mask framework, works with any differentiable model, yielding accurate, robust, and fast attributions that outperform existing algorithms on multiple XAI metrics.
The paper argues that feature attribution scores for generative language models lack a fixed meaning because each generated token is both output and input, leading to multiple distinct explanatory questions. It introduces the Attribution Contract framework, which explicitly defines the model score, fixed variables, target output, generation process, and eligible features, showing how these choices affect attribution outcomes. Experiments demonstrate that different contracts (e.g., local next-token vs. prompt-level) and model architectures (mixture-of-experts vs. masked-diffusion) yield markedly different attribution distributions, highlighting the need for careful contract specification.
The study evaluates 12 instruction‑tuned open‑weight LLMs on six causal‑graph benchmarks, testing five prompting strategies and four confidence sources. Findings show that LLMs tend to over‑predict edges, misclassify indirect or reversed edges as direct, and exhibit high over‑confidence, while conventional confidence estimates are unreliable and agreement signals offer limited improvement. The results suggest LLMs should be used as externally validated soft causal priors rather than definitive causal‑structure evidence.
arXiv:2606. 29171v1 Announce Type: cross Abstract: While existing data attribution methods can identify which training examples build specific mechanistic circuits, they cannot explain how training data shapes the high-level behavioral decisions a model learns to make.
arXiv:2504.00470v3 Announce Type: replace Abstract: To develop a trustworthy AI system, it is essential to understand its behavior using attribution methods, which aim to identify the input regions t...
arXiv:2606. 18383v1 Announce Type: new Abstract: Sparse autoencoders (SAEs) are increasingly used to extract interpretable features from language models (LMs), yet a central question remains: when can an SAE-based explanation be treated as a faithful view of an underlying frozen LM We study this through a post-hoc generalization framework that certifies the LM via a sparse proxy, obtained by replacing a native hidden activation with its pretrained SAE reconstruction.
The paper introduces MCIR-M, a global feature‑importance method that accounts for feature dependence by conditioning each feature on its strongly dependent neighbours and computing a normalized Mutual Correlation Impact Ratio. MCIR-M yields a population score between 0 and 1, reaching zero when a feature is exactly conditionally redundant, and can be estimated with only a fraction of the data. Experiments on synthetic redundancy setups and the UCI HAR benchmark show that MCIR-M produces more stable rankings under near‑duplicate predictors compared to existing methods such as SHAP, SAGE, HSIC, and MI‑based scores.
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:2606. 03885v1 Announce Type: new Abstract: Feature attribution methods explain predictions by assigning importance scores to input features.
arXiv:2608. 09209v1 Announce Type: cross Abstract: Neural language models trained on large crowdsourced corpora frequently exploit spurious surface patterns tied to target labels without true linguistic or causal relevance, boosting benchmark performance while failing on adversarial or out-of-distribution inputs.