mlsanityData to 14 Oct 2026

Data Curation Matters: Model Collapse and Spurious Shift Performance Prediction from Training on Uncurated Text Embeddings

arXiv 2506.17989 ↗First posted 22 Jun 2025cs.LGcited by 0

On the map: Synthetic data and self-improvement for LLMs → Model collapse from synthetic data

Abstract (arXiv)

Training models on uncurated Text Embeddings (TEs) derived from raw tabular data can lead to a severe failure mode known as model collapse, where predictions converge to a single class regardless of input. By comparing models trained with identical hyper-parameter configurations on both raw tabular data and their TE-derived counterparts, we find that collapse is a consistent failure mode in the latter setting. We introduce a set of metrics that capture the extent of model collapse, offering a new perspective on TE quality as a proxy for data curation. Our results reveal that TE alone does not effectively function as a curation layer - and that their quality significantly influences downstream learning. More insidiously, we observe that the presence of model collapse can yield artificially inflated and spurious Accuracy-on-the-Line correlation. These findings highlight the need for more nuanced curation and evaluation of embedding-based representations, particularly in out-of-distribution settings.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Model collapse from synthetic data, one of 13 directions in Synthetic data and self-improvement for LLMs. Builds on 33 of its 54 references that are in the map. First posted on arXiv 22 Jun 2025.

Builds on

33 of 54 references.

References · 33

  1. 2016XGBoost: A Scalable Tree Boosting Systemcited by 5,166
  2. 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,877
  3. 2019Invariant Risk Minimizationcited by 1,854
  4. 2021Revisiting Deep Learning Models for Tabular Datacited by 740
  5. 2019TabNet: Attentive Interpretable Tabular Learningcited by 648
  6. 2022MTEB: Massive Text Embedding Benchmarkcited by 640
  7. 2015Causal inference using invariant prediction: identification and confidence intervalscited by 536
  8. 2021Tabular Data: Deep Learning is Not All You Needcited by 535
  9. 2020Measuring Robustness to Natural Distribution Shifts in Image Classificationcited by 454
  10. 2021Retiring Adult: New Datasets for Fair Machine Learningcited by 449
  11. 2020TabTransformer: Tabular Data Modeling Using Contextual Embeddingscited by 409
  12. 2023A Comprehensive Survey on Test-Time Adaptation under Distribution Shiftscited by 359
21 more
  1. 2023Improving Text Embeddings with Large Language Modelscited by 303
  2. 2022TabLLM: Few-shot Classification of Tabular Data with Large Language Modelscited by 277
  3. 2021Accuracy on the Line: On the Strong Correlation Between Out-of-Distribution and In-Distribution Generalizationcited by 249
  4. 2016Statistics of Robust Optimization: A Generalized Empirical Likelihood Approachcited by 183
  5. 2021Well-tuned Simple Nets Excel on Tabular Datasetscited by 151
  6. 2022Agreement-on-the-Line: Predicting the Performance of Neural Networks under Distribution Shiftcited by 94
  7. 2020Transfer Learning for High-dimensional Linear Regression: Prediction, Estimation, and Minimax Optimalitycited by 90
  8. 2021Transfer Learning under High-dimensional Generalized Linear Modelscited by 70
  9. 2018Universal features of price formation in financial markets: perspectives from Deep Learningcited by 60
  10. 2020When is invariance useful in an Out-of-Distribution Generalization problem ?cited by 52
  11. 2023Improved Test-Time Adaptation for Domain Generalizationcited by 47
  12. 2023Diagnosing Model Performance Under Distribution Shiftcited by 39
  13. 2023TABLET: Learning From Instructions For Tabular Datacited by 28
  14. 2024Linq-Embed-Mistral Technical Reportcited by 20
  15. 2022Subgroup Robustness Grows On Trees: An Empirical Baseline Investigationcited by 17
  16. 2017Data-driven Optimal Cost Selection for Distributionally Robust Optimizationcited by 15
  17. 2023Unifying Distributionally Robust Optimization via Optimal Transport Theorycited by 13
  18. 2023Test-Time Adaptation Induces Stronger Accuracy and Agreement-on-the-Linecited by 11
  19. 2024Fully Test-time Adaptation for Tabular Datacited by 8
  20. 2024Reassessing the Validity of Spurious Correlations Benchmarkscited by 4
  21. 2024LLM Embeddings Improve Test-time Adaptation to Tabular Y|X-Shiftscited by 3

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Machine-generated text detection prevents language model collapse
  2. 2025EmbeddingGemma: Powerful and Lightweight Text Representations
  3. 2026Where Post-Training Quantization Breaks Text Embedders: A Measured Map Across Four Embedder Families
  4. 2024Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data
  5. 2024Beyond Model Collapse: Scaling Up with Synthesized Data Requires Verification
  6. 2025Layer by Layer: Uncovering Hidden Representations in Language Models
  7. 2024Recent advances in text embedding: A Comprehensive Review of Top-Performing Methods on the MTEB Benchmark
  8. 2026Break Step: Recursive Training Resonates with Replayed Sampling Noise