mlsanityData to 14 Oct 2026

Noiser: Bounded Input Perturbations for Attributing Large Language Models

arXiv 2504.02911 ↗First posted 3 Apr 2025cs.CL cs.AIcited by 4

Not on the map. Why?

Abstract (arXiv)

Feature attribution (FA) methods are common post-hoc approaches that explain how Large Language Models (LLMs) make predictions. Accordingly, generating faithful attributions that reflect the actual inner behavior of the model is crucial. In this paper, we introduce Noiser, a perturbation-based FA method that imposes bounded noise on each input embedding and measures the robustness of the model against partially noised input to obtain the input attributions. Additionally, we propose an answerability metric that employs an instructed judge model to assess the extent to which highly scored tokens suffice to recover the predicted output. Through a comprehensive evaluation across six LLMs and three tasks, we demonstrate that Noiser consistently outperforms existing gradient-based, attention-based, and perturbation-based FA methods in terms of both faithfulness and answerability, making it a robust and effective approach for explaining language model predictions.

Key facts. Cited by 4 papers in the map (data through 14 Oct 2026). Builds on 45 of its 57 references that are in the map. First posted on arXiv 3 Apr 2025.

Builds on

45 of 57 references.

References · 45

  1. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  2. 2017A Unified Approach to Interpreting Model Predictionscited by 6,555
  3. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,330
  4. 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,070
  5. 2017Axiomatic Attribution for Deep Networkscited by 3,384
  6. 2023Qwen Technical Reportcited by 3,344
  7. 2022Locating and Editing Factual Associations in GPTcited by 2,691
  8. 2013Visualizing and Understanding Convolutional Networkscited by 2,290
  9. 2017Learning Important Features Through Propagating Activation Differencescited by 1,458
  10. 2017Towards A Rigorous Science of Interpretable Machine Learningcited by 1,416
  11. 2023WizardLM: Empowering large pre-trained language models to follow complex instructionscited by 1,125
  12. 2023LIMA: Less Is More for Alignmentcited by 1,100
33 more
  1. 2023SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Modelscited by 810
  2. 2020Quantifying Attention Flow in Transformerscited by 775
  3. 2016The Mythos of Model Interpretabilitycited by 721
  4. 2023Instruction Tuning with GPT-4cited by 686
  5. 2019ERASER: A Benchmark to Evaluate Rationalized NLP Modelscited by 530
  6. 2018e-SNLI: Natural Language Inference with Natural Language Explanationscited by 468
  7. 2019Is Attention Interpretable?cited by 389
  8. 2019Explain Yourself! Leveraging Language Models for Commonsense Reasoningcited by 367
  9. 2016Rationalizing Neural Predictionscited by 281
  10. 2020A Diagnostic Study of Explainability Techniques for Text Classificationcited by 159
  11. 2023Self-Alignment with Instruction Backtranslationcited by 151
  12. 2024InFoBench: Evaluating Instruction Following Ability in Large Language Modelscited by 135
  13. 2019Interpretable Neural Predictions with Differentiable Binary Variablescited by 126
  14. 2020Learning to Faithfully Rationalize by Constructioncited by 113
  15. 2021The Out-of-Distribution Problem in Explainability and Search Methods for Feature Importance Explanationscited by 87
  16. 2024Long Is More for Alignment: A Simple but Tough-to-Beat Baseline for Instruction Fine-Tuningcited by 85
  17. 2020Aligning Faithful Interpretations with their Social Attributioncited by 79
  18. 2023Inseq: An Interpretability Toolkit for Sequence Generation Modelscited by 78
  19. 2016"What is Relevant in a Text Document?": An Interpretable Machine Learning Approachcited by 53
  20. 2023Sequential Integrated Gradients: a simple but effective method for explaining language modelscited by 52
  21. 2014Extraction of Salient Sentences from Labelled Documentscited by 50
  22. 2021Robust Explainability: A Tutorial on Gradient-Based Attribution Methods for Deep Neural Networkscited by 48
  23. 2022A Comparative Study of Faithfulness Metrics for Model Interpretability Methodscited by 39
  24. 2021UNIREX: A Unified Learning Framework for Language Model Rationale Extractioncited by 32
  25. 2021Evaluating the Faithfulness of Importance Measures in NLP by Recursively Masking Allegedly Important Tokens and Retrainingcited by 31
  26. 2021Rationales for Sequential Predictionscited by 26
  27. 2022An Empirical Study on Explanations in Out-of-Domain Settingscited by 21
  28. 2021Enjoy the Salience: Towards Better Transformer-based Faithful Explanations with Word Saliencecited by 16
  29. 2023Incorporating Attribution Importance for Improving Faithfulness Metricscited by 14
  30. 2022Saliency Map Verbalization: Comparing Feature Importance Representations from Model-free and Instruction-based Methodscited by 13
  31. 2024ReAGent: A Model-agnostic Feature Attribution Method for Generative Language Modelscited by 12
  32. 2023REFER: An End-to-end Rationale Extraction Framework for Explanation Regularizationcited by 4
  33. 2021Attention vs non-attention for a Shapley-based explanation methodcited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 2; 2026: 2.

4 papers.

  1. 2025Interpretation Meets Safety: A Survey on Interpretation Methods and Tools for Improving LLM Safetycited by 10
  2. 2025DePass: Unified Feature Attributing by Simple Decomposed Forward Passcited by 1
  3. 2026TRUE: A Trustworthy Unified Explanation Framework for Large Language Model Reasoningcited by 0
  4. 2026Enhancing Assessment of Self-Consistency in LLM Explanations using Perturbation Strengthcited by 0