mlsanityData to 14 Oct 2026

When and Why Test Generators for Deep Learning Produce Invalid Inputs: an Empirical Study

arXiv 2212.11368 ↗First posted 21 Dec 2022cs.SE cs.LGcited by 18

Not on the map. Why?

Abstract (arXiv)

Testing Deep Learning (DL) based systems inherently requires large and representative test sets to evaluate whether DL systems generalise beyond their training datasets. Diverse Test Input Generators (TIGs) have been proposed to produce artificial inputs that expose issues of the DL systems by triggering misbehaviours. Unfortunately, such generated inputs may be invalid, i.e., not recognisable as part of the input domain, thus providing an unreliable quality assessment. Automated validators can ease the burden of manually checking the validity of inputs for human testers, although input validity is a concept difficult to formalise and, thus, automate. In this paper, we investigate to what extent TIGs can generate valid inputs, according to both automated and human validators. We conduct a large empirical study, involving 2 different automated validators, 220 human assessors, 5 different TIGs and 3 classification tasks. Our results show that 84% artificially generated inputs are valid, according to automated validators, but their expected label is not always preserved. Automated validators reach a good consensus with humans (78% accuracy), but still have limitations when dealing with feature-rich datasets.

Key facts. Cited by 18 papers in the map (data through 14 Oct 2026). Builds on 26 of its 61 references that are in the map. First posted on arXiv 21 Dec 2022.

Builds on

26 of 61 references.

References · 26

  1. 2014Very Deep Convolutional Networks for Large-Scale Image Recognitioncited by 15,364
  2. 2014ImageNet Large Scale Visual Recognition Challengecited by 8,498
  3. 2017Fashion-MNIST: a Novel Image Dataset for Benchmarking Machine Learning Algorithmscited by 4,349
  4. 2013Auto-Encoding Variational Bayescited by 4,053
  5. 2014Deep Learning Face Attributes in the Wildcited by 3,693
  6. 2018Large Scale GAN Training for High Fidelity Natural Image Synthesiscited by 2,626
  7. 2014Striving for Simplicity: The All Convolutional Netcited by 990
  8. 2021Generalized Out-of-Distribution Detection: A Surveycited by 802
  9. 2017DeepXplore: Automated Whitebox Testing of Deep Learning Systemscited by 311
  10. 2017DeepTest: Automated Testing of Deep-Neural-Network-driven Autonomous Carscited by 251
  11. 2019Machine Learning Testing: Survey, Landscapes and Horizonscited by 218
  12. 2018DeepGauge: Multi-Granularity Testing Criteria for Deep Learning Systemscited by 147
14 more
  1. 2018Guiding Deep Learning System Testing using Surprise Adequacycited by 124
  2. 2019Taxonomy of Real Faults in Deep Learning Systemscited by 90
  3. 2018DLFuzz: Differential Fuzzing Testing of Deep Learning Systemscited by 65
  4. 2019Misbehaviour Prediction for Autonomous Driving Systemscited by 55
  5. 2020Model-based Exploration of the Frontier of Behaviours for Deep Learning System Testingcited by 40
  6. 2021DeepHyperion: Exploring the Feature Space of Deep Learning-Based Systems through Illumination Searchcited by 29
  7. 2020Towards Characterizing Adversarial Defects of Deep Learning Software from the Lens of Uncertaintycited by 26
  8. 2020Reducing DNN Labelling Cost using Surprise Adequacy: An Industrial Case Study for Autonomous Drivingcited by 22
  9. 2021Distribution-Aware Testing of Neural Networks Using Generative Modelscited by 22
  10. 2020SINVAD: Search-based Image Space Navigation for DNN Image Classifier Test Input Generationcited by 15
  11. 2021DeepMetis: Augmenting a Deep Learning Test Set to Increase its Mutation Scorecited by 12
  12. 2021Exposing Previously Undetectable Faults in Deep Neural Networkscited by 11
  13. 2022Black-box Safety Analysis and Retraining of DNNs based on Feature Extraction and Clusteringcited by 8
  14. 2019DeepSmartFuzzer: Reward Guided Test Generation For Deep Learningcited by 7

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 5; 2024: 7; 2025: 4; 2026: 2.

18 papers.

Sort:

By area:

  1. 2024Online Safety Analysis for LLMs: a Benchmark, an Assessment, and a Path Forwardcited by 6
  2. 2023Evaluating the Robustness of Test Selection Methods for Deep Neural Networkscited by 4
  3. 2023LUNA: A Model-Based Universal Analysis Framework for Large Language Modelscited by 4
  4. 2023Evaluating the Impact of Flaky Simulators on Testing Autonomous Driving Systemscited by 4
  5. 2024Bridging the Gap between Real-world and Synthetic Images for Testing Autonomous Driving Systemscited by 4
  6. 2024Benchmarking Generative AI Models for Deep Learning Test Input Generationcited by 4
  7. 2024Targeted Deep Learning System Boundary Testingcited by 3
  8. 2023GIST: Generated Inputs Sets Transferability in Deep Learningcited by 1
  9. 2024Can Search-Based Testing with Pareto Optimization Effectively Cover Failure-Revealing Test Inputs?cited by 1
  10. 2025Simulator Ensembles for Trustworthy Autonomous Driving Systems Testingcited by 1
  11. 2026TestifAI: Tomography-Based Testing for Deep Learning Systemscited by 1
  12. 2023Adopting Two Supervisors for Efficient Use of Large-Scale Remote Deep Neural Networkscited by 0
6 more
  1. 2024Generating Minimalist Adversarial Perturbations to Test Object-Detection Models: An Adaptive Multi-Metric Evolutionary Search Approachcited by 0
  2. 2024DiffGAN: A Test Generation Approach for Differential Testing of Deep Neural Networks for Image Analysiscited by 0
  3. 2025Generating Realistic, Diverse, and Fault-Revealing Inputs with Latent Space Interpolation for Testing Deep Neural Networkscited by 0
  4. 2025Towards Assessing Deep Learning Test Input Generatorscited by 0
  5. 2025Effort-Optimized, Accuracy-Driven Labelling and Validation of Test Inputs for DL Systems: A Mixed-Integer Linear Programming Approachcited by 0
  6. 2026HyperNet-Adaptation for Diffusion-Based Test Case Generationcited by 0