mlsanityData to 14 Oct 2026

Quantifying the Agreement Between Data-Influence and Data-Similarity to Understand LLM Behavior

arXiv 2606.23591 ↗First posted 22 Jun 2026cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

One way to understand LLM behavior is to trace its output back to the training data. Two types of measures are commonly used for output tracing: data-similarity and data-influence. The former is cheaper while the latter is believed to be more accurate. Even though many works have compared them for ground-truth tasks, no such comparisons exist for output tracing. Here, we fill this gap and precisely quantify the commonalities and differences between the two measures. We do this by first ranking the training documents according to each measure and then computing the overlap between the two rankings. Our main finding is that the two rankings agree significantly, but there is an asymmetry between them: The top documents of data-similarity are assigned more consistent ranks by data-influence than the other way around. This result is valid across a range of experiments involving OLMo2-1B, Qwen3-1.7B, LlaMa3.2-1B, Gemma3-1B, and GPT2. We exploit the asymmetry to obtain a favorable cost-accuracy trade-off by using the costly data-influence to refine the results of data-similarity.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 26 of its 36 references that are in the map. First posted on arXiv 22 Jun 2026.

Builds on

26 of 36 references.

References · 26

  1. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,626
  2. 2017Understanding Black-box Predictions via Influence Functionscited by 1,741
  3. 2020Estimating Training Data Influence by Tracing Gradient Descentcited by 580
  4. 2021Deep Learning on a Data Diet: Finding Important Examples Early in Trainingcited by 533
  5. 2024LESS: Selecting Influential Data for Targeted Instruction Tuningcited by 532
  6. 2024DataComp-LM: In search of the next generation of training sets for language modelscited by 398
  7. 20242 OLMo 2 Furiouscited by 340
  8. 2023Studying Large Language Model Generalization with Influence Functionscited by 326
  9. 2024Rethinking Machine Unlearning for Large Language Modelscited by 257
  10. 2021On Hallucination and Predictive Uncertainty in Conditional Language Generationcited by 176
  11. 2024Infini-gram: Scaling Unbounded n-gram Language Models to a Trillion Tokenscited by 153
  12. 2019On the Accuracy of Influence Functions for Measuring Group Effectscited by 147
14 more
  1. 2018Fast Approximate Natural Gradient Descent in a Kronecker-factored Eigenbasiscited by 146
  2. 2023Speak, Memory: An Archaeology of Books Known to ChatGPT/GPT-4cited by 125
  3. 2020FastIF: Scalable Influence Functions for Efficient Model Interpretation and Debuggingcited by 120
  4. 2023Copyright Violations and Large Language Modelscited by 108
  5. 2024What is Your Data Worth to GPT? LLM-Scale Data Valuation with Influence Functionscited by 94
  6. 2023Style Over Substance: Evaluation Biases for Large Language Modelscited by 70
  7. 2024Variational Learning is Effective for Large Deep Networkscited by 65
  8. 2024Scalable Influence and Fact Tracing for Large Language Model Pretrainingcited by 42
  9. 2023Self-Influence Guided Data Reweighting for Language Model Pre-trainingcited by 37
  10. 2025OLMoTrace: Tracing Language Model Outputs Back to Trillions of Training Tokenscited by 27
  11. 2023The Memory Perturbation Equation: Understanding Model's Sensitivity to Datacited by 16
  12. 2024In2Core: Leveraging Influence Functions for Coreset Selection in Instruction Finetuning of Large Language Modelscited by 9
  13. 2023A Simple and Efficient Baseline for Data Attribution on Imagescited by 5
  14. 2025Better Hessians Matter: Studying the Impact of Curvature Approximations in Influence Functionscited by 3

Built on by

Not cited yet.