mlsanityData to 14 Oct 2026

Can You Rely on Your Model Evaluation? Improving Model Evaluation with Synthetic Test Data

arXiv 2310.16524 ↗First posted 25 Oct 2023cs.LGcited by 32

Not on the map. Why?

Abstract (arXiv)

Evaluating the performance of machine learning models on diverse and underrepresented subgroups is essential for ensuring fairness and reliability in real-world applications. However, accurately assessing model performance becomes challenging due to two main issues: (1) a scarcity of test data, especially for small subgroups, and (2) possible distributional shifts in the model's deployment setting, which may not align with the available test data. In this work, we introduce 3S Testing, a deep generative modeling framework to facilitate model evaluation by generating synthetic test sets for small subgroups and simulating distributional shifts. Our experiments demonstrate that 3S Testing outperforms traditional baselines -- including real test data alone -- in estimating model performance on minority subgroups and under plausible distributional shifts. In addition, 3S offers intervals around its performance estimates, exhibiting superior coverage of the ground truth compared to existing approaches. Overall, these results raise the question of whether we need a paradigm shift away from limited real test data towards synthetic test data.

Key facts. Cited by 32 papers in the map (data through 14 Oct 2026). Builds on 54 of its 90 references that are in the map. First posted on arXiv 25 Oct 2023.

Builds on

54 of 90 references.

References · 54

  1. 2016Simple and Scalable Predictive Uncertainty Estimation using Deep Ensemblescited by 3,926
  2. 2019Benchmarking Neural Network Robustness to Common Corruptions and Perturbationscited by 2,877
  3. 2016Concrete Problems in AI Safetycited by 1,642
  4. 2019Do ImageNet Classifiers Generalize to ImageNet?cited by 1,515
  5. 2018Datasheets for Datasetscited by 1,386
  6. 2020WILDS: A Benchmark of in-the-Wild Distribution Shiftscited by 1,375
  7. 2020Beyond Accuracy: Behavioral Testing of NLP models with CheckListcited by 1,006
  8. 2019Modeling Tabular data using Conditional GANcited by 699
  9. 2021Tabular Data: Deep Learning is Not All You Needcited by 535
  10. 2020Underspecification Presents Challenges for Credibility in Modern Machine Learningcited by 496
  11. 2021Deep Neural Networks and Tabular Data: A Surveycited by 491
  12. 2017Data Augmentation Generative Adversarial Networkscited by 278
38 more
  1. 2018Pros and Cons of GAN Evaluation Measurescited by 224
  2. 2019Hidden Stratification Causes Clinically Meaningful Failures in Machine Learning for Medical Imagingcited by 216
  3. 2020HateCheck: Functional Tests for Hate Speech Detection Modelscited by 216
  4. 2021How Faithful is your Synthetic Data? Sample-level Metrics for Evaluating and Auditing Generative Modelscited by 196
  5. 2021A Fine-Grained Analysis on Distribution Shiftcited by 171
  6. 2019Learning from Synthetic Data for Crowd Counting in the Wildcited by 148
  7. 2022Leveraging Unlabeled Data to Predict Out-of-Distribution Performancecited by 145
  8. 2021Predicting with Confidence on Unseen Distributionscited by 116
  9. 2021DECAF: Generating Fair Synthetic Data Using Causally-Aware Generative Networkscited by 109
  10. 2016Generative Models and Model Criticism via Optimized Maximum Mean Discrepancycited by 100
  11. 2019FairVis: Visual Analytics for Discovering Intersectional Bias in Machine Learningcited by 93
  12. 2020Model Patching: Closing the Subgroup Performance Gap with Data Augmentationcited by 83
  13. 2021Mandoline: Model Evaluation under Distribution Shiftcited by 55
  14. 2020RealCause: Realistic Causal Inference Benchmarkingcited by 46
  15. 2021Discover the Unknown Biased Attribute of an Image Classifiercited by 46
  16. 2021Online Adaptation to Label Distribution Shiftcited by 44
  17. 2022Turning the Tables: Biased, Imbalanced, Dynamic Tabular Datasets for ML Evaluationcited by 40
  18. 2019Using Wasserstein Generative Adversarial Networks for the Design of Monte Carlo Simulationscited by 32
  19. 2023Beyond Privacy: Navigating the Opportunities and Challenges of Synthetic Datacited by 29
  20. 2023Synthetic data, real errors: how (not) to publish and use synthetic datacited by 29
  21. 2015A Test of Relative Similarity For Model Selection in Generative Modelscited by 25
  22. 2021What can Data-Centric AI Learn from Data and ML Engineering?cited by 24
  23. 2019Characterizing Bias in Classifiers using Generative Modelscited by 23
  24. 2018Learning Tasks for Multitask Learning: Heterogenous Patient Populations in the ICUcited by 22
  25. 2019Tutorial: Safe and Reliable Machine Learningcited by 22
  26. 2022Data-IQ: Characterizing subgroups with heterogeneous outcomes in tabular datacited by 22
  27. 2022Validating Causal Inference Methodscited by 20
  28. 2020Empirical Frequentist Coverage of Deep Learning Uncertainty Quantification Procedurescited by 19
  29. 2021Conditional Synthetic Data Generation for Robust Machine Learning Applications with Limited Pandemic Datacited by 18
  30. 2017Synthetic Data for Social Goodcited by 15
  31. 2022Conditional Generation of Medical Time Series for Extrapolation to Underrepresented Populationscited by 14
  32. 2022Data-SUITE: Data-centric identification of in-distribution incongruous examplescited by 12
  33. 2018Generating Photo-Realistic Training Data to Improve Face Recognition Accuracycited by 11
  34. 2021The Medkit-Learn(ing) Environment: Medical Decision Modelling through Simulationcited by 11
  35. 2022Effect of Balancing Data Using Synthetic Data on the Performance of Machine Learning Classifiers for Intrusion Detection in Computer Networkscited by 11
  36. 2021Simulated Adversarial Testing of Face Recognition Modelscited by 10
  37. 2022A Human-Centric Take on Model Monitoringcited by 9
  38. 2021Model-based metrics: Sample-efficient estimates of predictive model subpopulation performancecited by 7

50 most cited of 54 All 54 (JSON)

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 1; 2024: 12; 2025: 10; 2026: 9.

32 papers.

Sort:

By area:

  1. 2024Why Tabular Foundation Models Should Be a Research Prioritycited by 106
  2. 2024AutoEval Done Right: Using Synthetic Data for Model Evaluationcited by 57
  3. 2025TIMER: Temporal Instruction Modeling and Evaluation for Longitudinal Clinical Recordscited by 34
  4. 2023RadEdit: stress-testing biomedical vision models via diffusion image editingcited by 28
  5. 2024ClavaDDPM: Multi-relational Data Synthesis with Cluster-guided Diffusion Modelscited by 28
  6. 2025Position: All Current Generative Fidelity and Diversity Metrics are Flawedcited by 12
  7. 2024A structured regression approach for evaluating model performance across intersectional subgroupscited by 9
  8. 2024Muffin or Chihuahua? Challenging Multimodal Large Language Models with Multipanel VQAcited by 9
  9. 2025TAI3: Testing Agent Integrity in Interpreting User Intentcited by 9
  10. 2024LaTable: Towards Large Tabular Modelscited by 8
  11. 2024Fill In The Gaps: Model Calibration and Generalization with Synthetic Datacited by 8
  12. 2024Context-Aware Testing: A New Paradigm for Model Testing with Large Language Modelscited by 8
20 more
  1. 2024FairJob: A Real-World Dataset for Fairness in Online Systemscited by 6
  2. 2026Reasoning-Driven Synthetic Data Generation and Evaluationcited by 5
  3. 2025A Design-based Solution for Causal Inference with Text: Can a Language Model Be Too Large?cited by 3
  4. 2024A Bias-Variance Decomposition for Ensembles over Multiple Synthetic Datasetscited by 2
  5. 2026MIDST Challenge at SaTML 2025: Membership Inference over Diffusion-models-based Synthetic Tabular datacited by 2
  6. 2024Generative Conditional Distributions by Neural (Entropic) Optimal Transportcited by 1
  7. 2024Visual Car Brand Classification by Implementing a Synthetic Image Dataset Creation Pipelinecited by 0
  8. 2025B-RIGHT: Benchmark Re-evaluation for Integrity in Generalized Human-Object Interaction Testingcited by 0
  9. 2025Physical foundations for trustworthy medical imaging: a review for artificial intelligence researcherscited by 0
  10. 2025TRUST: Test-time Resource Utilization for Superior Trustworthinesscited by 0
  11. 2025Beyond Internal Data: Constructing Complete Datasets for Fairness Testingcited by 0
  12. 2025Beyond Internal Data: Bounding and Estimating Fairness from Incomplete Datacited by 0
  13. 2025Using Synthetic Data to estimate the True Error is theoretically and practically doablecited by 0
  14. 2026Generating synthetic electronic health record data using agent-based models to evaluate machine learning robustness under mass casualty incidentscited by 0
  15. 2026Demographically-Conditioned Synthetic Medical Images for Bias Mitigation and Bias Detection in Disease Classifierscited by 0
  16. 2026FUSE: Feature-Wise Unified Specialization with Cross-Column Exchange for Mixed-Type Tabular Flow Matchingcited by 0
  17. 2026Position: Fairness Failure in Generative Models is an Evaluation Problemcited by 0
  18. 2026IDSPACE: A Novel Document Generator for Reliable Evaluation of Digital Identity Verification Systems [Extended Technical Report]cited by 0
  19. 2026Domain generalization and synthetic data in object detection: the enabler, the probe, and the gapcited by 0
  20. 2026Can Synthetic UAV Benchmarks Be Trusted? Real-Synthetic Validation for UAV Model Selectioncited by 0