mlsanityData to 14 Oct 2026

Understanding Dataset Difficulty with V-Usable Information

arXiv 2110.08420 ↗First posted 16 Oct 2021cs.CL cs.AI cs.LGcited by 291

Not on the map. Why?

Cited by 142 in LLM alignment / RLHF / preference optimisation, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Estimating the difficulty of a dataset typically involves comparing state-of-the-art models to humans; the bigger the performance gap, the harder the dataset is said to be. However, this comparison provides little understanding of how difficult each instance in a given distribution is, or what attributes make the dataset difficult for a given model. To address these questions, we frame dataset difficulty -- w.r.t. a model V -- as the lack of V-textitusable information (Xu et al., 2019), where a lower value indicates a more difficult dataset for V. We further introduce textitpointwise V-information (PVI) for measuring the difficulty of individual instances w.r.t. a given distribution. While standard evaluation metrics typically only compare different models for the same dataset, V-textitusable information and PVI also permit the converse: for a given model V, we can compare different datasets, as well as different instances/slices of the same dataset. Furthermore, our framework allows for the interpretability of different input attributes via transformations of the input, which we use to discover annotation artefacts in widely-used NLP benchmarks.

Key facts. Cited by 291 papers in the map (data through 14 Oct 2026). Builds on 39 of its 67 references that are in the map. First posted on arXiv 16 Oct 2021.

Builds on

39 of 67 references.

References · 39

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,399
  2. 2019RoBERTa: A Robustly Optimized BERT Pretraining Approachcited by 13,372
  3. 2019BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehensioncited by 5,893
  4. 2019DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lightercited by 4,307
  5. 2017A Broad-Coverage Challenge Corpus for Sentence Understanding through Inferencecited by 2,682
  6. 2015A large annotated corpus for learning natural language inferencecited by 1,847
  7. 2017Understanding Black-box Predictions via Influence Functionscited by 1,741
  8. 2019Do ImageNet Classifiers Generalize to ImageNet?cited by 1,515
  9. 2017Explanation in Artificial Intelligence: Insights from the Social Sciencescited by 1,206
  10. 2018Co-teaching: Robust Training of Deep Neural Networks with Extremely Noisy Labelscited by 1,059
  11. 2018Neural Network Acceptability Judgmentscited by 958
  12. 2020Language (Technology) is Power: A Critical Survey of "Bias" in NLPcited by 956
27 more
  1. 2017Automated Hate Speech Detection and the Problem of Offensive Languagecited by 736
  2. 2018An Empirical Study of Example Forgetting during Deep Neural Network Learningcited by 678
  3. 2018Annotation Artifacts in Natural Language Inference Datacited by 671
  4. 2019Data Shapley: Equitable Valuation of Data for Machine Learningcited by 629
  5. 2020Dataset Cartography: Mapping and Diagnosing Datasets with Training Dynamicscited by 425
  6. 2020Fine-Tuning Pretrained Language Models: Weight Initializations, Data Orders, and Early Stoppingcited by 384
  7. 2019Unsupervised Label Noise Modeling and Loss Correctioncited by 328
  8. 2019Verified Uncertainty Calibrationcited by 309
  9. 2018Hypothesis Only Baselines in Natural Language Inferencecited by 295
  10. 2019Towards Efficient Data Valuation Based on the Shapley Valuecited by 281
  11. 2020Identifying Mislabeled Data using the Area Under the Margin Rankingcited by 261
  12. 2020On the Stability of Fine-tuning BERT: Misconceptions, Explanations, and Strong Baselinescited by 255
  13. 2019Build it Break it Fix it for Dialogue Safety: Robustness from Adversarial Human Attackcited by 176
  14. 2019Show Your Work: Improved Reporting of Experimental Resultscited by 169
  15. 2020A Theory of Usable Information Under Computational Constraintscited by 162
  16. 2020Adversarial Filters of Dataset Biasescited by 160
  17. 2017Active Bias: Training More Accurate Neural Networks by Emphasizing High Variance Samplescited by 147
  18. 2021Competency Problems: On Finding and Removing Artifacts in Language Datacited by 91
  19. 2021Conditional probing: measuring usable information beyond a baselinecited by 62
  20. 2018What Makes Reading Comprehension Questions Easier?cited by 60
  21. 2021Dynaboard: An Evaluation-As-A-Service Platform for Holistic Next-Generation Benchmarkingcited by 53
  22. 2021What Context Features Can Transformer Language Models Use?cited by 50
  23. 2020What do Models Learn from Question Answering Datasets?cited by 45
  24. 2021Combining Feature and Instance Attribution to Detect Artifactscited by 38
  25. 2021Rissanen Data Analysis: Examining Dataset Characteristics via Description Lengthcited by 22
  26. 2021A Bayesian Framework for Information-Theoretic Probingcited by 22
  27. 2018Are All Training Examples Created Equal? An Empirical Studycited by 16

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2022: 5; 2023: 56; 2024: 111; 2025: 71; 2026: 48.

291 papers.

Sort:

By area:

  1. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  2. 2023Qwen Technical Reportcited by 3,344
  3. 2023AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Modelscited by 785
  4. 2023OpenAssistant Conversations -- Democratizing Large Language Model Alignmentcited by 721
  5. 2023DataComp: In search of the next generation of multimodal datasetscited by 644
  6. 2023RLAIF vs. RLHF: Scaling Reinforcement Learning from Human Feedback with AI Feedbackcited by 625
  7. 2024RewardBench: Evaluating Reward Models for Language Modelingcited by 499
  8. 2023LMSYS-Chat-1M: A Large-Scale Real-World LLM Conversation Datasetcited by 461
  9. 2023UltraFeedback: Boosting Language Models with Scaled AI Feedbackcited by 406
  10. 2024Interpretable Preferences via Multi-Objective Reward Modeling and Mixture-of-Expertscited by 376
  11. 2024Skywork-Reward: Bag of Tricks for Reward Modeling in LLMscited by 327
  12. 2024JudgeBench: A Benchmark for Evaluating LLM-based Judgescited by 318
38 more
  1. 2023Evaluating Large Language Models at Evaluating Instruction Followingcited by 302
  2. 2024RLHF Workflow: From Reward Modeling to Online RLHFcited by 267
  3. 2024KTO: Model Alignment as Prospect Theoretic Optimizationcited by 264
  4. 2024HelpSteer2: Open-source dataset for training top-performing reward modelscited by 236
  5. 2023FLASK: Fine-grained Language Model Evaluation based on Alignment Skill Setscited by 201
  6. 2023Generative Judge for Evaluating Alignmentcited by 169
  7. 2024Continual Learning for Large Language Models: A Surveycited by 147
  8. 2023HelpSteer: Multi-attribute Helpfulness Dataset for SteerLMcited by 142
  9. 2025OpenRubrics: Towards Scalable Synthetic Rubric Generation for Reward Modeling and LLM Alignmentcited by 132
  10. 2024HelpSteer2-Preference: Complementing Ratings with Preferencescited by 130
  11. 2024Taming Overconfidence in LLMs: Reward Calibration in RLHFcited by 112
  12. 2023On the Diversity and Realism of Distilled Dataset: An Efficient Dataset Distillation Paradigmcited by 109
  13. 2024Unpacking DPO and PPO: Disentangling Best Practices for Learning from Preference Feedbackcited by 107
  14. 2024ARGS: Alignment as Reward-Guided Searchcited by 103
  15. 2025GuardReasoner: Towards Reasoning-based LLM Safeguardscited by 100
  16. 2024How to Evaluate Reward Models for RLHFcited by 88
  17. 2023Anchor Points: Benchmarking Models with Much Fewer Examplescited by 86
  18. 2023Do Models Explain Themselves? Counterfactual Simulatability of Natural Language Explanationscited by 85
  19. 2023Cognitive Overload: Jailbreaking Large Language Models with Overloaded Logical Thinkingcited by 81
  20. 2022Efficient Methods for Natural Language Processing: A Surveycited by 78
  21. 2023ReCEval: Evaluating Reasoning Chains via Correctness and Informativenesscited by 77
  22. 2024RRM: Robust Reward Model Training Mitigates Reward Hackingcited by 76
  23. 2025HelpSteer3-Preference: Open Human-Annotated Preference Data across Diverse Tasks and Languagescited by 75
  24. 2024Foundational Autoraters: Taming Large Language Models for Better Automatic Evaluationcited by 74
  25. 2025InternLM-XComposer2.5-Reward: A Simple Yet Effective Multi-Modal Reward Modelcited by 74
  26. 2024Clio: Privacy-Preserving Insights into Real-World AI Usecited by 72
  27. 2024Self-Generated Critiques Boost Reward Modeling for Language Modelscited by 71
  28. 2024Towards Analyzing and Understanding the Limitations of DPO: A Theoretical Perspectivecited by 69
  29. 2024Rejection Improves Reliability: Training LLMs to Refuse Unknown Questions Using RL from Knowledge Feedbackcited by 66
  30. 2024MJ-Bench: Is Your Multimodal Reward Model Really a Good Judge for Text-to-Image Generation?cited by 66
  31. 2024How Large Language Models Encode Context Knowledge? A Layer-Wise Probing Studycited by 64
  32. 2024Aligning Large Language Models with Representation Editing: A Control Perspectivecited by 64
  33. 2024Quantile Regression for Distributional Reward Models in RLHFcited by 60
  34. 2024DailyDilemmas: Revealing Value Preferences of LLMs with Quandaries of Daily Lifecited by 59
  35. 2023Uncertainty-Penalized Reinforcement Learning from Human Feedback with Diverse Reward LoRA Ensemblescited by 56
  36. 2023SALMON: Self-Alignment with Instructable Reward Modelscited by 55
  37. 2024Towards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neuronscited by 53
  38. 2024Online Iterative Reinforcement Learning from Human Feedback with General Preference Modelcited by 52

50 most cited of 291 All 291 (JSON)