mlsanityData to 14 Oct 2026

Grounding Deep Learning Framework Testing in Data Flow

arXiv 2610.15808 ↗First posted 11 Oct 2026cs.SE cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Data flow underlies deep learning (DL) framework execution, including automatic differentiation and operator-fusion optimization. Bugs involving data flow can cause incorrect results or unexpected behavior. Existing testing methods represent models primarily by operator topology, omitting runtime tensor values and properties and operator input requirements. They therefore cannot distinguish data flows in models with similar topologies, even when these differences affect framework behavior. We introduce the Tensor Flow Graph (TFG) to capture runtime tensor states and operator input requirements. Its nodes represent operators and their input requirements, while its edges record tensor values and properties. Building on the TFG, we propose FlowMuT, a mutation-based testing method grounded in data flow. Mutations are represented as constrained subgraph replacements, with mutation operators derived from prior work and an analysis of mutable data flow objects. In each round, FlowMuT builds a TFG for the seed model, pairs mutation operators with TFG sites, and filters illegal candidates using data flow constraints. It uses LinUCB to select a candidate based on coverage and impact, executes the mutant, updates selection feedback, and uses each accepted mutant as the next seed. We evaluate FlowMuT on PyTorch and MindSpore using representative models from diverse application scenarios. FlowMuT submitted 55 bug reports for cases missed by all baseline methods; developers confirmed 41 reports and fixed the bugs described in 7 reports. FlowMuT achieves 43.4% higher data flow diversity than the strongest baseline and a legal mutant rate of 96.7%, compared with 86.4% without constraints. These results show that testing from a data flow perspective is practical and effective.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 39 of its 64 references that are in the map. First posted on arXiv 11 Oct 2026.

Builds on

39 of 64 references.

References · 39

  1. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,305
  2. 2021SegFormer: Simple and Efficient Design for Semantic Segmentation with Transformerscited by 2,940
  3. 2021Masked-attention Mask Transformer for Universal Image Segmentationcited by 2,351
  4. 2021SwinIR: Image Restoration Using Swin Transformercited by 1,887
  5. 2021Swin Transformer V2: Scaling Up Capacity and Resolutioncited by 1,228
  6. 2021EfficientNetV2: Smaller Models and Faster Trainingcited by 1,135
  7. 2023ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoderscited by 712
  8. 2024Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inferencecited by 554
  9. 2024YOLOv10: Real-Time End-to-End Object Detectioncited by 515
  10. 2019Machine Learning Testing: Survey, Landscapes and Horizonscited by 218
  11. 2023RTMPose: Real-Time Multi-Person Pose Estimation based on MMPosecited by 175
  12. 2023EfficientAD: Accurate Visual Anomaly Detection at Millisecond-Level Latenciescited by 160
27 more
  1. 2022Scene Text Recognition with Permuted Autoregressive Sequence Modelscited by 149
  2. 2024RT-DETRv2: Improved Baseline with Bag-of-Freebies for Real-Time Detection Transformercited by 91
  3. 2018Glow: Graph Lowering Compiler Techniques for Neural Networkscited by 75
  4. 2019Automatic Testing and Improvement of Machine Translationcited by 36
  5. 2022NNSmith: Generating Diverse and Valid Test Cases for Deep Learning Compilerscited by 30
  6. 2022Free Lunch for Testing: Fuzzing Deep-Learning Libraries from Open Sourcecited by 28
  7. 2018Intel nGraph: An Intermediate Representation, Compiler, and Executor for Deep Learningcited by 23
  8. 2021DocTer: Documentation Guided Fuzzing for Testing Deep Learning API Functionscited by 22
  9. 2023Fuzzing Automatic Differentiation in Deep-Learning Librariescited by 17
  10. 2022Muffin: Testing Deep Learning Libraries via Neural Architecture Fuzzingcited by 16
  11. 2022Coverage-Guided Tensor Compiler Fuzzing with Joint IR-Pass Mutationcited by 10
  12. 2022Fuzzing Deep Learning Compilers with HirGencited by 7
  13. 2023NeuRI: Diversifying DNN Generation via Inductive Rule Inferencecited by 7
  14. 2024Deep Learning Library Testing: Definition, Methods and Challengescited by 5
  15. 2024Enhancing Differential Testing With LLMs For Testing Deep Learning Librariescited by 5
  16. 2022COMET: Coverage-guided Model Generation For Deep Learning Library Testingcited by 4
  17. 2022IvySyn: Automated Vulnerability Discovery in Deep Learning Frameworkscited by 4
  18. 2023TorchProbe: Fuzzing Dynamic Deep Learning Compilerscited by 4
  19. 2025Understanding LLM-Centric Challenges for Deep Learning Frameworks: An Empirical Analysiscited by 4
  20. 2025Improving Deep Learning Framework Testing with Model-Level Metamorphic Testingcited by 3
  21. 2025Deep Learning Framework Testing via Model Mutation: How Far Are We?cited by 2
  22. 2024MoCo: Fuzzing Deep Learning Libraries via Assembling Codecited by 1
  23. 2024Mutation-Based Deep Learning Framework Testing Method in JavaScript Environmentcited by 1
  24. 2024Subgraph-Oriented Testing for Deep Learning Librariescited by 1
  25. 2025Deep Learning Framework Testing via Heuristic Guidance Based on Multiple Model Measurementscited by 1
  26. 2025Scalpel: Automotive Deep Learning Framework Testing via Assembling Model Componentscited by 1
  27. 2026Testing Deep Learning Libraries via Neurosymbolic Constraint Learningcited by 1

Built on by

Not cited yet.