mlsanityData to 14 Oct 2026

Enhancing LLM Robustness to Perturbed Instructions: An Empirical Study

arXiv 2504.02733 ↗First posted 3 Apr 2025cs.CLcited by 17

Not on the map. Why?

Abstract (arXiv)

Large Language Models (LLMs) are highly vulnerable to input perturbations, as even a small prompt change may result in a substantially different output. Existing methods to enhance LLM robustness are primarily focused on perturbed data samples, whereas improving resiliency to perturbations of task-level instructions has remained relatively underexplored. In this work, we focus on character- and word-level edits of task-specific instructions, which substantially degrade downstream performance. We experiment with a variety of techniques to enhance the robustness of LLMs, including self-denoising and representation alignment, testing different models (Llama 3 and Flan-T5), datasets (CoLA, QNLI, SST-2) and instructions (both task-oriented and role-oriented). We find that, on average, self-denoising -- whether performed by a frozen LLM or a fine-tuned model -- achieves substantially higher performance gains than alternative strategies, including more complex baselines such as ensembling and supervised methods.

Key facts. Cited by 17 papers in the map (data through 14 Oct 2026). Builds on 36 of its 49 references that are in the map. First posted on arXiv 3 Apr 2025.

Builds on

36 of 49 references.

References · 36

  1. 2021LoRA: Low-Rank Adaptation of Large Language Modelscited by 14,380
  2. 2020A Simple Framework for Contrastive Learning of Visual Representationscited by 12,049
  3. 2018GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understandingcited by 4,899
  4. 2016SQuAD: 100,000+ Questions for Machine Comprehension of Textcited by 4,148
  5. 2023QLoRA: Efficient Finetuning of Quantized LLMscited by 3,418
  6. 2022Scaling Instruction-Finetuned Language Modelscited by 3,024
  7. 2017A Broad-Coverage Challenge Corpus for Sentence Understanding through Inferencecited by 2,682
  8. 2019Certified Adversarial Robustness via Randomized Smoothingcited by 1,340
  9. 2020Beyond Accuracy: Behavioral Testing of NLP models with CheckListcited by 1,006
  10. 2018Neural Network Acceptability Judgmentscited by 958
  11. 2019Is BERT Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailmentcited by 671
  12. 2023Capabilities of GPT-4 on Medical Challenge Problemscited by 638
24 more
  1. 2020BERT-ATTACK: Adversarial Attack Against BERT Using BERTcited by 491
  2. 2018TextBugger: Generating Adversarial Text Against Real-world Applicationscited by 381
  3. 2018Black-box Generation of Adversarial Text Sequences to Evade Deep Learning Classifierscited by 345
  4. 2023State of What Art? A Call for Multi-Prompt LLM Evaluationcited by 310
  5. 2023Improving Text Embeddings with Large Language Modelscited by 303
  6. 2018Adversarial Example Generation with Syntactically Controlled Paraphrase Networkscited by 286
  7. 2021Adversarial GLUE: A Multi-Task Benchmark for Robustness Evaluation of Language Modelscited by 232
  8. 2018Stress Test Evaluation for Natural Language Inferencecited by 213
  9. 2023On the Robustness of ChatGPT: An Adversarial and Out-of-distribution Perspectivecited by 211
  10. 2024SecAlign: Defending Against Prompt Injection with Preference Optimizationcited by 210
  11. 2023Defending Large Language Models Against Jailbreaking Attacks Through Goal Prioritizationcited by 170
  12. 2021Measure and Improve Robustness in NLP Models: A Surveycited by 107
  13. 2021Evaluating the Robustness of Neural Language Models to Input Perturbationscited by 87
  14. 2023Evaluating the Zero-shot Robustness of Instruction-tuned Language Modelscited by 66
  15. 2023PromptBench: A Unified Library for Evaluation of Large Language Modelscited by 44
  16. 2022SemAttack: Natural Textual Attacks via Different Semantic Spacescited by 42
  17. 2022Robustness of Learning from Task Instructionscited by 35
  18. 2023Text-CRS: A Generalized Certified Robustness Framework against Textual Adversarial Attackscited by 34
  19. 2024Resilience of Large Language Models for Noisy Instructionscited by 20
  20. 2024Prompt Perturbation in Retrieval-Augmented Generation based Large Language Modelscited by 18
  21. 2023Turbulence: Systematically and Automatically Testing Instruction-Tuned Large Language Models for Codecited by 9
  22. 2023How (not) to ensemble LVLMs for VQAcited by 4
  23. 2024MaskPure: Improving Defense Against Text Adversaries with Stochastic Purificationcited by 2
  24. 2024From Test-Taking to Test-Making: Examining LLM Authoring of Commonsense Assessment Itemscited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 5; 2026: 12.

17 papers.

Sort:

By area:

  1. 2026AgentNoiseBench: Benchmarking Robustness of Tool-Using LLM Agents Under Noisy Conditioncited by 23
  2. 2026Brittlebench: Quantifying LLM robustness via prompt sensitivitycited by 21
  3. 2025Robustness in Large Language Models: A Survey of Mitigation Strategies and Evaluation Metricscited by 12
  4. 2026Learning to Act under Noise: Enhancing Agent Robustness via Noisy Environmentscited by 6
  5. 2025Truth or Twist? Optimal Model Selection for Reliable Label Flipping Evaluation in LLM-based Counterfactualscited by 5
  6. 2026Evaluating Robustness of Large Language Models in Enterprise Applications: Benchmarks for Perturbation Consistency Across Formats and Languagescited by 5
  7. 2026FragileFlow: Spectral Control of Correct-but-Fragile Predictions for Foundation Model Robustnesscited by 3
  8. 2025Conversational Speech Reveals Structural Robustness Failures in SpeechLLM Backbonescited by 2
  9. 2025Fine-Tuning on Noisy Instructions: Effects on Generalization and Performancecited by 2
  10. 2025Perceive and Calibrate: Analyzing and Enhancing Robustness of Medical Multi-Modal Large Language Modelscited by 2
  11. 2026Where Do Prompt Perturbations Break Generation? A Segment-Level View of Robustness in LoRA-Tuned Language Modelscited by 1
  12. 2026Beyond Confidence: The Rhythms of Reasoning in Generative Modelscited by 0
5 more
  1. 2026The Text Uncanny Valley: Non-Monotonic Performance Degradation in LLM Information Retrievalcited by 0
  2. 2026Emergent retokenization symmetry in large language models: phenomenology and applicationscited by 0
  3. 2026Prompt Perturbation for Reliable LLM Evaluation over Comparison Graphscited by 0
  4. 2026Test-Time Augmentation for LLMs: When Input Diversity Beats Output Diversity at Matched Computecited by 0
  5. 2026Prompt-Robust Language Models: Which Training Strategies Work?cited by 0