mlsanityData to 7 Oct 2026

Feedback-guided Data Synthesis for Imbalanced Classification

arXiv 2310.00158 ↗First posted 29 Sep 2023cs.CV cs.AI cs.LGcited by 33

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

Abstract (arXiv)

Current status quo in machine learning is to use static datasets of real images for training, which often come from long-tailed distributions. With the recent advances in generative models, researchers have started augmenting these static datasets with synthetic data, reporting moderate performance improvements on classification tasks. We hypothesize that these performance gains are limited by the lack of feedback from the classifier to the generative model, which would promote the usefulness of the generated samples to improve the classifier's performance. In this work, we introduce a framework for augmenting static datasets with useful synthetic samples, which leverages one-shot feedback from the classifier to drive the sampling of the generative model. In order for the framework to be effective, we find that the samples must be close to the support of the real data of the task at hand, and be sufficiently diverse. We validate three feedback criteria on a long-tailed dataset (ImageNet-LT) as well as a group-imbalanced dataset (NICO++). On ImageNet-LT, we achieve state-of-the-art results, with over 4 percent improvement on underrepresented classes while being twice efficient in terms of the number of generated synthetic samples. NICO++ also enjoys marked boosts of over 5 percent in worst group accuracy. With these results, our framework paves the path towards effectively leveraging state-of-the-art text-to-image models as data sources that can be queried to improve downstream applications.

Key facts. Cited by 33 papers in the map (data through 7 Oct 2026). Placed in Model collapse from synthetic data, one of 13 directions in Synthetic data and self-improvement for LLMs. Builds on 81 of its 95 references that are in the map. First posted on arXiv 29 Sep 2023.

Builds on

81 of 95 references.

References · 81

  1. 2015Deep Residual Learning for Image Recognitioncited by 51,809
  2. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  3. 2020Denoising Diffusion Probabilistic Modelscited by 20,851
  4. 2021High-Resolution Image Synthesis with Latent Diffusion Modelscited by 18,694
  5. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  6. 2020Denoising Diffusion Implicit Modelscited by 9,655
  7. 2021Diffusion Models Beat GANs on Image Synthesiscited by 8,185
  8. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,005
  9. 2022Hierarchical Text-Conditional Image Generation with CLIP Latentscited by 6,539
  10. 2022Photorealistic Text-to-Image Diffusion Models with Deep Language Understandingcited by 6,322
  11. 2022Classifier-Free Diffusion Guidancecited by 5,717
  12. 2017mixup: Beyond Empirical Risk Minimizationcited by 5,155
38 more
  1. 2021Zero-Shot Text-to-Image Generationcited by 4,276
  2. 2019Generative Modeling by Estimating Gradients of the Data Distributioncited by 4,067
  3. 2021The Power of Scale for Parameter-Efficient Prompt Tuningcited by 3,769
  4. 2021GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Modelscited by 3,595
  5. 2019Invariant Risk Minimizationcited by 1,832
  6. 2023Consistency Modelscited by 1,778
  7. 2019Class-Balanced Loss Based on Effective Number of Samplescited by 1,239
  8. 2019Distributionally Robust Neural Networks for Group Shifts: On the Importance of Regularization for Worst-Case Generalizationcited by 1,232
  9. 2023ImageReward: Learning and Evaluating Human Preferences for Text-to-Image Generationcited by 1,010
  10. 2019Learning Imbalanced Datasets with Label-Distribution-Aware Margin Losscited by 987
  11. 2022eDiff-I: Text-to-Image Diffusion Models with an Ensemble of Expert Denoiserscited by 921
  12. 2016Playing for Data: Ground Truth from Computer Gamescited by 816
  13. 2017Deep Bayesian Active Learning with Image Datacited by 812
  14. 2019Decoupling Representation and Classifier for Long-Tailed Recognitioncited by 806
  15. 2019Large-Scale Long-Tailed Recognition in an Open Worldcited by 704
  16. 2022Automatic Chain of Thought Prompting in Large Language Modelscited by 684
  17. 2021Just Train Twice: Improving Group Robustness without Training Group Informationcited by 522
  18. 2022Beyond neural scaling laws: beating power law scaling via data pruningcited by 519
  19. 2023Scaling up GANs for Text-to-Image Synthesiscited by 492
  20. 2017VisDA: The Visual Domain Adaptation Challengecited by 432
  21. 2020Reliable Fidelity and Diversity Metrics for Generative Modelscited by 425
  22. 2023TIFA: Accurate and Interpretable Text-to-Image Faithfulness Evaluation with Question Answeringcited by 416
  23. 2020Balanced Meta-Softmax for Long-Tailed Visual Recognitioncited by 397
  24. 2023Synthetic Data from Diffusion Models Improves ImageNet Classificationcited by 326
  25. 2022Is synthetic data from generative models ready for image recognition?cited by 320
  26. 2023Your Diffusion Model is Secretly a Zero-Shot Classifiercited by 301
  27. 2020Gradient Starvation: A Learning Proclivity in Neural Networkscited by 262
  28. 2022Easily Accessible Text-to-Image Generation Amplifies Demographic Stereotypes at Large Scalecited by 262
  29. 2023Stable Bias: Analyzing Societal Representations in Diffusion Modelscited by 225
  30. 2021Parametric Contrastive Learningcited by 217
  31. 2021Improving Calibration for Long-Tailed Recognitioncited by 212
  32. 2017GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibriumcited by 208
  33. 2022Improving Out-of-Distribution Robustness via Selective Augmentationcited by 202
  34. 2021Simple data balancing achieves competitive worst-group-accuracycited by 177
  35. 2023StableRep: Synthetic Images from Text-to-Image Models Make Strong Visual Representation Learnerscited by 177
  36. 2022Fake it till you make it: Learning transferable representations from synthetic ImageNet clonescited by 171
  37. 2017Cost-Effective Active Learning for Deep Image Classificationcited by 165
  38. 2023DiffuMask: Synthesizing Images with Pixel-level Annotations for Semantic Segmentation Using Diffusion Modelscited by 162

50 most cited of 81 All 81 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 3; 2024: 10; 2025: 15; 2026: 5.

33 papers.

Sort:

By area:

  1. 2024An Introduction to Vision-Language Modelingcited by 175
  2. 2024CinePile: A Long Video Question Answering Dataset and Benchmarkcited by 112
  3. 2024Beyond Model Collapse: Scaling Up with Synthesized Data Requires Verificationcited by 43
  4. 2024Consistency-diversity-realism Pareto fronts of conditional image generative modelscited by 35
  5. 2024Self-Consuming Generative Models with Curated Data Provably Optimize Human Preferencescited by 35
  6. 2026Inference-time Physics Alignment of Video Generative Models with Latent World Modelscited by 32
  7. 2024Improving Geo-diversity of Generated Images with Contextualized Vendi Score Guidancecited by 24
  8. 2025Improving the Scaling Laws of Synthetic Data with Deliberate Practicecited by 17
  9. 2024The Pitfalls of Memorization: When Memorization Hurts Generalizationcited by 16
  10. 2023Gen2Det: Generate to Detectcited by 12
  11. 2023GeNIe: Generative Hard Negative Images Through Diffusioncited by 8
  12. 2025Why Less is More (Sometimes): A Theory of Data Curationcited by 8
21 more
  1. 2023Unlocking Pre-trained Image Backbones for Semantic Image Synthesiscited by 6
  2. 2024Diffusion Curriculum: Synthetic-to-Real Data Curriculum via Image-Guided Diffusioncited by 6
  3. 2025Increasing the Utility of Synthetic Images through Chamfer Guidancecited by 6
  4. 2025The Intricate Dance of Prompt Complexity, Quality, Diversity, and Consistency in T2I Modelscited by 6
  5. 2024Position: Cracking the Code of Cascading Disparity Towards Marginalized Communitiescited by 4
  6. 2025DIMCIM: A Quantitative Evaluation Framework for Default-mode Diversity and Generalization in Text-to-Image Generative Modelscited by 4
  7. 2025Letting Trajectories Spread: Quality-Preserving Control for Diverse Flow Matchingcited by 3
  8. 2025Spend Wisely: Maximizing Post-Training Gains in Iterative Synthetic Data Bootstrappingcited by 2
  9. 2025Gaussian Splatting is an Effective Data Generator for 3D Object Detectioncited by 2
  10. 2025LoRA-Loop: Closing the Synthetic Replay Cycle for Continual VLM Learningcited by 2
  11. 2025Filtering with Confidence: When Data Augmentation Meets Conformal Predictioncited by 2
  12. 2026Learnability-Guided Diffusion for Dataset Distillationcited by 2
  13. 2025LoFT: LoRA-fused Training Dataset Generation with Few-shot Guidancecited by 1
  14. 2025SoK: Can Synthetic Images Replace Real Data? A Survey of Utility and Privacy of Synthetic Image Generationcited by 1
  15. 2024LoGex: Improved tail detection of extremely rare histopathology classes via guided diffusioncited by 0
  16. 2025Boosting Statistic Learning with Synthetic Data from Pretrained Large Modelscited by 0
  17. 2025Beyond Objects: Contextual Synthetic Data Generation for Fine-Grained Classificationcited by 0
  18. 2025HiGFA: Hierarchical Guidance for Fine-grained Data Augmentation with Diffusion Modelscited by 0
  19. 2026LiBaGS: Lightweight Boundary Gap Synthesis for Targeted Synthetic Data Selectioncited by 0
  20. 2026Personalized Generative Models for Contextual Debiasingcited by 0
  21. 2026Post-Generation Curation of Synthetic Images via Homogeneous-Heterogeneous Splittingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Efficient Long-Tail Learning in Latent Space by sampling Synthetic Data
  2. 2025Recursive Learning Without Collapse: A Weighting-Based Stabilization Framework
  3. 2023Real-Fake: Effective Training Data Synthesis Through Distribution Matching
  4. 2023On the Stability of Iterative Retraining of Generative Models on their own Data
  5. 2023Training on Thin Air: Improve Image Classification with Generated Data
  6. 2023Diversify, Don't Fine-Tune: Scaling Up Visual Recognition Training with Synthetic Images
  7. 2025Bias-Corrected Data Synthesis for Imbalanced Learning
  8. 2023Generative Oversampling for Imbalanced Data via Majority-Guided VAE