mlsanityData to 7 Oct 2026

Model collapse from synthetic data

85 papers0 this weekHigh confidence in this grouping: the papers form one clear topicHigh confidence in this grouping: the papers form one clear topic · About

Keywords: collapse, synthetic, training, time, generative

Key facts. Model collapse from synthetic data: 85 papers on arXiv since 2023 (data through 7 Oct 2026). 28 of its 85 mapped papers (33%) cite “The Curse of Recursion: Training on Generated Data Makes Models Forget” (2023, arXiv 2305.17493). 26 of its 85 mapped papers (31%) cite “A Tale of Tails: Model Collapse as a Change of Scaling Laws” (2024, arXiv 2402.07043). 7 new papers 17 Sep–7 Oct.

What’s new

17 Sep – 7 Oct 2026

7 new papers, 17 Sep – 7 Oct; 0 this week.

  1. Earlier · 7
  2. From Behavior to Provenance: Attributing Tabular Foundation Models to Synthetic Pretraining Data also fits: Self-rewarding alignment and self-training
  3. No Model Required: Text Entropy Rate Filtering Mitigates Iterative Fine-Tuning Collapse also fits: LLM synthetic data generation
  4. Feature Selective Model Collapse in Diffusion Models: Total Replacement versus Fixed-Budget Training
  5. Effective Synthetic Data Curation Requires Group-Level Signals also fits: LLM synthetic data generation
  6. Source Identification Is Not Fitness Testing: Measuring the Limits of Synthetic-Data Attribution also fits: LLM synthetic data generation
  7. Shared Autoregressive Context Can Distort Relationships in Synthetic Data also fits: LLM synthetic data generation
  8. Let Training Guide Selection: Online Synthetic Data Filtering via Real-Anchored Utility also fits: LLM synthetic data generation

Foundations

Most cited by papers here

  1. The Curse of Recursion: Training on Generated Data Makes Models Forget2023cited by 28 here
  2. A Tale of Tails: Model Collapse as a Change of Scaling Laws2024cited by 26 here
  3. Model Collapse Demystified: The Case of Regression2024cited by 22 here
  4. Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Data2024cited by 22 here
  5. How Bad is Training on Synthetic Data? A Statistical Analysis of Language Model Collapse2024cited by 18 here

Lineage

The papers this direction builds on, oldest first.

Each paper on the line cites the one before.

  1. 2022Large Language Models Can Self-Improvecited by 4 here
  2. 2022Self-Instruct: Aligning Language Models with Self-Generated Instructionscited by 4 here
  3. 2024A Tale of Tails: Model Collapse as a Change of Scaling Lawscited by 26 here
  4. 2024Model Collapse Demystified: The Case of Regressioncited by 22 here
  5. 2024How Bad is Training on Synthetic Data? A Statistical Analysis of Language Model Collapsecited by 18 here
  6. 2024Strong Model Collapsecited by 17 here
  7. 2024How to Synthesize Text Data without Model Collapse?cited by 8 here
  8. 2025Position: Model Collapse Does Not Mean What You Thinkcited by 10 here
  9. 2025A Closer Look at Model Collapse: From a Generalization-to-Memorization Perspectivecited by 3 here
15 more it builds on, not in this chain
  1. 2022STaR: Bootstrapping Reasoning With Reasoningcited by 3 here
  2. 2023The Curse of Recursion: Training on Generated Data Makes Models Forgetcited by 28 here
  3. 2023Reinforced Self-Training (ReST) for Language Modelingcited by 2 here
  4. 2023Feedback-guided Data Synthesis for Imbalanced Classificationcited by 2 here
  5. 2024Self-Play Fine-Tuning Converts Weak Language Models to Strong Language Modelscited by 3 here
  6. 2024Is Model Collapse Inevitable? Breaking the Curse of Recursion by Accumulating Real and Synthetic Datacited by 22 here
  7. 2024Beyond Model Collapse: Scaling Up with Synthesized Data Requires Verificationcited by 13 here
  8. 2024Regurgitative Training: The Value of Real Data in Training Large Language Modelscited by 4 here
  9. 2024Model Collapse in the Self-Consuming Chain of Diffusion Finetuning: A Novel Perspective from Quantitative Trait Modelingcited by 3 here
  10. 2024Analyzing and Mitigating Model Collapse in Rectified Flow Modelscited by 3 here
  11. 2024Rate of Model Collapse in Recursive Trainingcited by 6 here
  12. 2025Recursive Learning Without Collapse: A Weighting-Based Stabilization Frameworkcited by 5 here
  13. 2025A Theoretical Perspective: How to Prevent Model Collapse in Self-consuming Training Loopscited by 3 here
  14. 2025When Models Don't Collapse: On the Consistency of Iterative MLEcited by 4 here
  15. 2025Preventing Model Collapse Under Overparametrization: Optimal Mixing Ratios for Interpolation Learning and Ridge Regressioncited by 3 here

The 24 most cited of 66.

Browse

30 of 85, ranked 6–35

  1. 6Strong Model Collapse202417
  2. 7Beyond Model Collapse: Scaling Up with Synthesized Data Requires Verification202413
  3. 8Position: Model Collapse Does Not Mean What You Think202510
  4. 9Rate of Model Collapse in Recursive Training20246
  5. 10Recursive Learning Without Collapse: A Weighting-Based Stabilization Framework20255
  6. 11When Models Don't Collapse: On the Consistency of Iterative MLE20254
  7. 12A Theoretical Perspective: How to Prevent Model Collapse in Self-consuming Training Loops20253
  8. 13A Closer Look at Model Collapse: From a Generalization-to-Memorization Perspective20253
  9. 14Preventing Model Collapse Under Overparametrization: Optimal Mixing Ratios for Interpolation Learning and Ridge Regression20253
  10. 15Model Collapse in the Self-Consuming Chain of Diffusion Finetuning: A Novel Perspective from Quantitative Trait Modeling20243
All 30
  1. 16Analyzing and Mitigating Model Collapse in Rectified Flow Models20243
  2. 17Machine-generated text detection prevents language model collapse20252
  3. 18Empowering Time Series Analysis with Synthetic Data: A Survey and Outlook in the Era of Foundation Models20252
  4. 19A Probabilistic Perspective on Model Collapse20252
  5. 20LLM Web Dynamics: Tracing Model Collapse in a Network of LLMs20252
  6. 21ChatTS: Aligning Time Series with LLMs via Synthetic Data for Enhanced Understanding and Reasoning20242
  7. 22Feedback-guided Data Synthesis for Imbalanced Classification20232
  8. 23Position: the Stochastic Parrot in the Coal Mine. Model Collapse is a Threat to Low-Resource Communities20261
  9. 24When Does Model Collapse Occur in Structured Interactive Learning?20261
  10. 25Multi-modal Synthetic Data Training and Model Collapse: Insights from VLMs and Diffusion Models20251
  11. 26Escaping Model Collapse via Synthetic Data Verification: Near-term Improvements and Long-term Convergence20251
  12. 27The Butterfly Effect of Model Editing: Few Edits Can Trigger Large Language Models Collapse20241
  13. 28Context Collapse: In-Context Learning and Model Collapse20260
  14. 29SIGMA: Scalable Spectral Insights for LLM Model Collapse20260
  15. 30From Collapse to Improvement: Statistical Perspectives on the Evolutionary Dynamics of Iterative Training on Contaminated Sources20260
  16. 31Quantifying Error Propagation and Model Collapse in Diffusion Models20260
  17. 32A Markovian View of Iterative-Feedback Loops in Image Generative Models: Neural Resonance and Model Collapse20260
  18. 33When and Where to Reset Matters for Long-Term Test-Time Adaptation20260
  19. 34Language Generation with Replay: A Learning-Theoretic View of Model Collapse20260
  20. 35Dynamic Context Evolution for Scalable Synthetic Data Generation20260