Pith. sign in

REVIEW 3 major objections 5 minor 38 references

Generating Highly Structured Test Inputs Leveraging Constraint-Guided Graph Refinement

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Pith's one-line read GRAphRef proposes that structured test inputs can be unified as graphs, so one mutation engine plus constraint repair can generate valid, semantics-preserving tests for 3D meshes and other modalities.

desk verdict An honest, well-structured research proposal for graph-based structured input generation, but with no evaluation and an overclaimed semantic-validity guarantee. read the letter →

arxiv 2507.21271 v1 pith:YWJTLJNS submitted 2025-07-28 cs.SE

classification cs.SE
keywords graph-basedtestinputgenerationfuzztestingstructuredinputs3Dmeshpointcloudsconstraint-guidedrefinementneighbor-awaremutationsemanticvalidity
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper is trying to establish that test input generation for highly structured data—3D meshes, point clouds, images, and text—can be unified through graphs. It proposes GRAphRef, which converts each input type into a graph, mutates the graph with 20 operators guided by local neighborhood similarity, then repairs constraint violations such as degenerate triangles and non-manifold edges. The authors claim this yields inputs that are both syntactically and semantically valid, avoiding the waste of discarded invalid inputs that afflicts byte-level fuzzers. If the claim holds, one reusable tool could replace handcrafted generators for many structured domains. The paper lays out the framework and a planned eight-system confirmatory study, but does not yet report experimental results.

What carries the argument

The carrying mechanism is the graph representation with three components: a converter that turns images, meshes, point clouds, and text into typed graphs; a neighbor-aware mutator that biases 20 mutation operators toward locally related vertices or edges; and a constraint-guided refiner that detects violations of declarative constraints (face orientation, area thresholds, edge connectivity, manifoldness) and repairs them with per-violation fix patterns. The neighbor-similarity bias is what is supposed to turn arbitrary perturbation into semantically coherent variation, while the refiner is what recovers validity so inputs are not thrown away.

What would settle it

Run the planned ablation with neighbor-aware mutation disabled: if the no-neighbor version preserves downstream predictions just as often as the full version on the eight mesh systems, the neighbor-similarity hypothesis is not supported.

Watch

Extended reading notes

Core claim

The central claim is that structural constraints and semantic meaning live in the same graph object: format, connectivity, and geometric validity can be expressed as graph constraints, and semantics can be preserved by mutating neighbor-similar regions together. GRAphRef encodes constraints in a DSL extended from ISLa, checks them after each mutation, and applies predefined repairs such as merging duplicate vertices, removing degenerate triangles, and limiting edge additions to triangles. The paper hypothesizes that this design produces a high valid-input rate, preserves downstream model predictions, and adds only linear overhead. It plans to test these hypotheses by comparing GRAphRef against AFL, Saffron, MeshAttack, and two ablated versions on eight mesh-processing AI systems.

Load-bearing premise

The whole approach depends on the assumption that parts of an input that are close together in the graph are also close in meaning, so mutating them together keeps the input's semantics intact; the paper has not yet tested this assumption.

Editorial extensions

If this is right

  • If GRAphRef works as claimed, fuzzing a new structured input type reduces to writing a graph converter and constraint set instead of a new grammar or generator.
  • Constraint-guided refinement could eliminate much of the 65% of fuzzing time that the paper says AFL wastes rejecting invalid meshes.
  • Neighbor-aware mutation, if it preserves semantics, would let testers generate large batches of valid variants that keep the subject AI model's behavior comparable to the original input.
  • The same engine could be shared across 3D mesh, point cloud, image, and text tasks, since each is expressed through the same graph API.
  • The planned ablations without refinement and without neighbor-aware mutation are designed to isolate each component's contribution to validity and semantic preservation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • A natural extension the paper does not pursue is to treat constraint-guided refinement as a standalone repair layer that any mutation-based fuzzer could call, decoupling semantic mutation from validity repair.
  • The paper's generalizability claim for images and text is not directly tested by its planned evaluation, which uses only mesh-processing systems; demonstrating the grid-graph and sequence-graph cases would require separate experiments.
  • If the neighbor-similarity principle proves out for meshes, the same principle could apply to any modality where local graph structure mirrors semantic units, such as molecular graphs, program abstract syntax trees, or audio spectrograms.
  • The semantic preservation metric (same top-1 prediction) is a coarse proxy; a sharper test would look at whether internal feature activations or gradient directions stay stable, which the paper does not propose.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes GRAphRef, a graph-based test input generation framework for highly structured inputs such as 3D meshes, point clouds, images, and text. The core idea is to convert structured inputs into graphs, apply neighbor-aware mutation operators that bias changes toward locally similar vertices/edges, and then enforce domain constraints via a constraint-guided refinement phase that repairs invalid inputs. The paper formulates four research questions with hypotheses, describes the framework's components (graph converter, neighbor-aware mutation engine, constraint-guided refinement), and presents an execution plan to evaluate against five baselines on eight mesh-processing AI systems. No experimental results are reported; the conclusion explicitly states that the evaluation is forthcoming. The central claim is that constraint-guided refinement ensures both syntactic and semantic validity of generated inputs.

Significance. If the framework delivered on its claims, it would address a genuine problem in testing AI systems that consume structured inputs: existing fuzzers produce many invalid inputs, and domain-specific generators do not generalize. The graph-based unification is a plausible and potentially valuable direction, and the paper's explicit hypotheses and controlled-variable design provide a reasonable template for future evaluation. However, the paper's significance cannot be assessed because the evaluation is absent, and the semantic-validity guarantee is unsupported by the constraint design. The paper's strengths are the clear problem statement and the separation of concerns (conversion, mutation, refinement), but these are not enough to establish the claimed contribution without empirical evidence.

major comments (3)
  1. [Section III, first paragraph; Section II, last paragraph] The claim that constraint-guided refinement "ensures both syntactic and semantic validity" is unsupported and, as stated, false. The constraints in Listing 1 (norm.z>0, area()>epsilon, connected_face()==1/2, fan_connected()==true) are purely geometric/topological; they do not reference downstream task semantics. Yet the paper's own SPS metric (Section V) defines semantic preservation as top-1 prediction consistency. No mechanism in the DSL or the refinement phase connects constraints to model behavior. Furthermore, the repair operations in Table I (merging duplicate vertices, removing degenerate triangles) alter geometry and can flip a model's prediction. The planned evaluation (Section VI) cannot verify this guarantee as stated; the claim should be weakened to structural/syntactic validity, or the framework must include task-aware constraints.
  2. [Section VI, Execution Plan; Section IX, Conclusion] The paper reports no experimental results. Section VI is an execution plan, not an evaluation; Section IX says "The forthcoming evaluation will assess its effectiveness." Consequently, hypotheses H1-H4 (Section IV) are untested. In particular, H3 is the load-bearing premise for semantic preservation: the assertion that mutations applied to locally related vertices/edges yield semantically coherent variants is plausible but has no supporting evidence. For a journal submission, a plan and hypotheses do not substantiate the paper's claims about validity and semantic preservation.
  3. [Section I, last bullet; Section VII, Time Limit] The motivating statistic that "65% of the time spent using AFL to generate meshes for MeshCNN was wasted on rejected meshes" is presented without methodology, datasets, or runtime details, so it cannot be verified. Similarly, Section VII's Time Limit bullet refers to observations "for S1-S5" to justify the 30-minute limit, but S1-S5 are never defined and no data are reported. These empirical assertions are inconsistent with the absence of an evaluation section; they should be substantiated with a pilot-study description or removed.
minor comments (5)
  1. [Throughout] The name "GRA PHREF" is rendered with a space in several places (e.g., in the abstract and Section III); the paper should use a consistent name, e.g., "GRAphRef."
  2. [Section VI.B] The text says "five baselines" but the list includes AFL, GRAPH GEN, GRAPH NONEIGHBOR, Saffron, and MeshAttack, which is five tools; however, the two ablated variants are GRAphRef configurations, not external baselines, so the wording is slightly confusing.
  3. [Figure 3] The grammar is labeled "partial," but the full grammar and the semantics of predicates such as fan_connected() and connected_face() are not defined; this makes it difficult to judge the expressiveness of the constraint DSL.
  4. [Section VII] The Threats to Validity section is written as if experiments were already conducted (e.g., "we did not see any increase in new types of errors with a higher time limit"), which conflicts with the stated "forthcoming evaluation." This should be reconciled.
  5. [Section V, SPS definition; Table II] The Semantic Preservation Score is defined as the percentage of mutated inputs with the same top-1 prediction label, but several evaluation subjects (MeshSDF, Point2Mesh, MeshWalker) perform regression or reconstruction tasks that do not have top-1 labels; the metric needs to be adapted for those models.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's claims are empirical hypotheses to be evaluated against external baselines, not derived from fitted parameters or self-citations.

full rationale

GRAphRef is a study-plan paper: it proposes a graph-based mutation and constraint-refinement framework and states hypotheses (H1a, H1b, H2, H3, H4) that are to be tested against external baselines (AFL, MeshAttack, Saffron) and ablated variants. No parameter is fitted to data and then renamed as a prediction; no result is derived from its own definition. The constraint grammar (Figure 3, Listing 1) encodes syntactic and geometric invariants such as face orientation, area thresholds, and manifoldness, while semantic preservation is defined independently as downstream prediction consistency (SPS) and is the subject of an explicitly stated, untested hypothesis (H3). The assertion that 'constraint-guided refinement ensures both syntactic and semantic validity' is an unsupported design claim, and the reported 65% AFL overhead statistic lacks methodology, but these are evidentiary weaknesses rather than circular reductions: the paper does not define syntactic or semantic validity in terms of GRAphRef's own outputs. There are no load-bearing self-citations; the cited ISLa work is external prior work. Consequently, the derivation chain is not circular, and the appropriate circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 1 invented entities

The paper's central design rests on untested domain assumptions about graph abstraction and semantic locality.

assumptions (4)
  • domain assumption Many structured inputs share common graph structures captured by extensible APIs (Hypothesis H1a).
    Stated as a hypothesis; the graph converter abstraction presupposes this unification is possible.
  • domain assumption Applying similar mutations to structurally similar neighbors preserves input semantics (Hypothesis H3).
    Underpins neighbor-aware mutation; not verified.
  • domain assumption Constraints such as manifoldness and connectivity can be repaired without destroying the mutation's intent (Hypothesis H2).
    Underpins constraint-guided refinement; not verified.
  • domain assumption A graph representation can capture the structure of meshes, images, point clouds, and text with adequate fidelity.
    Section III Graph Converter; the representation choice is manual and may lose information.
invented entities (1)
  • Constraint DSL (extended ISLa for graphs)
    purpose: Encodes structural constraints for refinement
    This is the framework's own notation, not independently validated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generating Highly Structured Test Inputs Leveraging Constraint-Guided Graph Refinement." pith.science (2026). https://pith.science/paper/YWJTLJNS

@misc{pith2026250721271,
  author       = {Pith},
  title        = {Pith review of: Generating Highly Structured Test Inputs Leveraging Constraint-Guided Graph Refinement},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YWJTLJNS}},
  note         = {Machine review of arXiv:2507.21271}
}
read the original abstract

[Context] Modern AI applications increasingly process highly structured data, such as 3D meshes and point clouds, where test input generation must preserve both structural and semantic validity. However, existing fuzzing tools and input generators are typically handcrafted for specific input types and often generate invalid inputs that are subsequently discarded, leading to inefficiency and poor generalizability. [Objective] This study investigates whether test inputs for structured domains can be unified through a graph-based representation, enabling general, reusable mutation strategies while enforcing structural constraints. We will evaluate the effectiveness of this approach in enhancing input validity and semantic preservation across eight AI systems. [Method] We develop and evaluate GRAphRef, a graph-based test input generation framework that supports constraint-based mutation and refinement. GRAphRef maps structured inputs to graphs, applies neighbor-similarity-guided mutations, and uses a constraint-refinement phase to repair invalid inputs. We will conduct a confirmatory study across eight real-world mesh-processing AI systems, comparing GRAphRef with AFL, MeshAttack, Saffron, and two ablated variants. Evaluation metrics include structural validity, semantic preservation (via prediction consistency), and performance overhead. Experimental data is derived from ShapeNetCore mesh seeds and model outputs from systems like MeshCNN and HodgeNet. Statistical analysis and component latency breakdowns will be used to assess each hypothesis.

Figures

Figures reproduced from arXiv: 2507.21271 by the authors.

Figure 2
Figure 2. Graph Representations of Different Data Types: (A) Image as a grid graph where pixels are vertices and edges connect neighbors. (B) Social network with users as vertices and friendships as edges. (C) Text as a sequential graph with characters as vertices and edges for order. (D) Triangle mesh with 3D points as vertices and edges forming the mesh. represent words and edges connect adjacent words. As another example, … view at source ↗
Figure 4
Figure 4. Working Example. (A) Ant-shaped mesh and eye region; (B) Extracted constraints; (C) Constructed graph of eye region; (D) Mutation result with vertex shift; (E) Graph after constraint-guided refinement. constraint-guided refinement. An overview is shown in Fig￾ure 1. Graph Converter. GRAPHREF converts structured data into graphs with semantic attributes. It supports four struc￾tural families: sequences (text/audio), … view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 22 canonical work pages

  1. [1]

    Pan-cancer computational histopathology reveals mutations, tumor composition and prognosis,

    Y . Fu, A. W. Jung, R. V . Torne, S. Gonzalez, H. Vöhringer, A. Shmatko, L. R. Yates, M. Jimenez-Linan, L. Moore, and M. Gerstung, “Pan-cancer computational histopathology reveals mutations, tumor composition and prognosis,” Nature cancer, vol. 1, no. 8, pp. 800–810, 2020

  2. [2]

    Apolloauto,

    Baidu, “Apolloauto,” 2024. [Online]. Available: https://github.com/ ApolloAuto/apollo

  3. [3]

    Meshcnn: a network with an edge,

    R. Hanocka, A. Hertz, N. Fish, R. Giryes, S. Fleishman, and D. Cohen- Or, “Meshcnn: a network with an edge,” ACM Transactions on Graphics (ToG), vol. 38, no. 4, pp. 1–12, 2019

  4. [4]

    Testing of autonomous driving systems: where are we and where should we go?

    G. Lou, Y . Deng, X. Zheng, M. Zhang, and T. Zhang, “Testing of autonomous driving systems: where are we and where should we go?” in Proceedings of the 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering, 2022, pp. 31–43

  5. [5]

    An open- source machine-learning application for predicting pixel-to-pixel ndvi regression from rgb calibrated images,

    L. Moscovini, L. Ortenzi, F. Pallottino, S. Figorilli, S. Violino, C. Pane, V . Capparella, S. Vasta, and C. Costa, “An open- source machine-learning application for predicting pixel-to-pixel ndvi regression from rgb calibrated images,” Computers and Electronics in Agriculture , vol. 216, p. 108536, 2024. [Online]. Available: https://www.sciencedirect.com...

  6. [6]

    Omniseg3d: Omniversal 3d segmentation via hierarchical contrastive learning,

    H. Ying, Y . Yin, J. Zhang, F. Wang, T. Yu, R. Huang, and L. Fang, “Omniseg3d: Omniversal 3d segmentation via hierarchical contrastive learning,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , 2024, pp. 20 612–20 622

  7. [7]

    American fuzzy lop

    Michal Zalewski, “American fuzzy lop.” [Online]. Available: https: //lcamtuf.coredump.cx/afl/

  8. [8]

    Saffron: Adaptive grammar-based fuzzing for worst-case analysis,

    X.-B. D. Le, C. Pasareanu, R. Padhye, D. Lo, W. Visser, and K. Sen, “Saffron: Adaptive grammar-based fuzzing for worst-case analysis,” SIGSOFT Softw. Eng. Notes , vol. 44, no. 4, p. 14, Dec. 2019. [Online]. Available: https://doi.org/10.1145/3364452.3364455

Show all 38 references
  1. [9]

    Open3D: A modern library for 3D data processing,

    Q.-Y . Zhou, J. Park, and V . Koltun, “Open3D: A modern library for 3D data processing,” arXiv:1801.09847, 2018

  2. [10]

    [Online]

    OpenCV , 2024. [Online]. Available: https://opencv.org/

  3. [11]

    Deepsdf: Learning continuous signed distance functions for shape rep- resentation,

    J. J. Park, P. Florence, J. Straub, R. Newcombe, and S. Lovegrove, “Deepsdf: Learning continuous signed distance functions for shape rep- resentation,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2019, pp. 165–174

  4. [12]

    Cfd vision 2030 study: A path to revolutionary computational aerosciences,

    “Cfd vision 2030 study: A path to revolutionary computational aerosciences,” 2025. [Online]. Available: https://ntrs.nasa.gov/citations/ 20140003093

  5. [13]

    3d adversarial attacks beyond point cloud,

    J. Zhang, L. Chen, B. Liu, B. Ouyang, Q. Xie, J. Zhu, W. Li, and Y . Meng, “3d adversarial attacks beyond point cloud,” Information Sciences, 2023

  6. [14]

    Black box fairness testing of machine learning models,

    A. Aggarwal, P. Lohia, S. Nagar, K. Dey, and D. Saha, “Black box fairness testing of machine learning models,” in Proceedings of the 2019 27th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering , ser. ESEC/FSE...

  7. [16]

    Input invariants,

    D. Steinhöfel and A. Zeller, “Input invariants,” in Proceedings of the 30th ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering , ser. ESEC/FSE 2022. New York, NY , USA: Association for Computing Machinery, 2022, p. 583–5...

  8. [17]

    FAUST: Dataset and evaluation for 3D mesh registration,

    F. Bogo, J. Romero, M. Loper, and M. J. Black, “FAUST: Dataset and evaluation for 3D mesh registration,” in Proceedings IEEE Conf. on Computer Vision and Pattern Recognition (CVPR) . Piscataway, NJ, USA: IEEE, Jun. 2014

  9. [18]

    ShapeNet: An Information-Rich 3D Model Repository,

    A. X. Chang, T. Funkhouser, L. Guibas, P. Hanrahan, Q. Huang, Z. Li, S. Savarese, M. Savva, S. Song, H. Su, J. Xiao, L. Yi, and F. Yu, “ShapeNet: An Information-Rich 3D Model Repository,” Stanford University — Princeton University — Toyota Technological Institute at Chicago, T...

  10. [19]

    Hodgenet: Learning spectral geometry on triangle meshes,

    D. Smirnov and J. Solomon, “Hodgenet: Learning spectral geometry on triangle meshes,” 2021. [Online]. Available: https://arxiv.org/abs/2104. 12826

  11. [20]

    3d shapenets: A deep representation for volumetric shapes,

    Z. Wu, S. Song, A. Khosla, F. Yu, L. Zhang, X. Tang, and J. Xiao, “3d shapenets: A deep representation for volumetric shapes,” 2015. [Online]. Available: https://arxiv.org/abs/1406.5670

  12. [21]

    Meshsdf: Differentiable iso-surface extraction,

    E. Remelli, A. Lukoianov, S. R. Richter, B. Guillard, T. Bagautdinov, P. Baque, and P. Fua, “Meshsdf: Differentiable iso-surface extraction,”

  13. [22]

    Point2mesh: a self-prior for deformable meshes,

    R. Hanocka, G. Metzer, R. Giryes, and D. Cohen-Or, “Point2mesh: a self-prior for deformable meshes,” ACM Transactions on Graphics , vol. 39, no. 4, Aug. 2020. [Online]. Available: http://dx.doi.org/10. 1145/3386569.3392415

  14. [23]

    Active co-analysis of a set of shapes,

    Y . Wang, S. Asafi, O. van Kaick, H. Zhang, D. Cohen-Or, and B. Chen, “Active co-analysis of a set of shapes,” ACM Trans. Graph. , vol. 31, no. 6, Nov. 2012. [Online]. Available: https://doi.org/10.1145/2366145.2366184

  15. [24]

    Meshwalker: Deep mesh understanding by random walks,

    A. Lahav and A. Tal, “Meshwalker: Deep mesh understanding by random walks,” 2020. [Online]. Available: https://arxiv.org/abs/2006. 05353

  16. [25]

    3D menagerie: Modeling the 3D shape and pose of animals,

    S. Zuffi, A. Kanazawa, D. Jacobs, and M. J. Black, “3D menagerie: Modeling the 3D shape and pose of animals,” in IEEE Conf. on Computer Vision and Pattern Recognition (CVPR) , Jul. 2017

  17. [26]

    Deepgcns: Can gcns go as deep as cnns?

    G. Li, M. Müller, A. Thabet, and B. Ghanem, “Deepgcns: Can gcns go as deep as cnns?” 2019. [Online]. Available: https://arxiv.org/abs/1904.03751

  18. [27]

    Boundary-aware geometric encoding for semantic segmentation of point clouds,

    “Boundary-aware geometric encoding for semantic segmentation of point clouds,” vol. 35. [Online]. Available: https://ojs.aaai.org/index. php/AAAI/article/view/16232

  19. [28]

    Human3.6m: Large scale datasets and predictive methods for 3d human sensing in natural environments,

    C. Ionescu, D. Papava, V . Olaru, and C. Sminchisescu, “Human3.6m: Large scale datasets and predictive methods for 3d human sensing in natural environments,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 36, no. 7, pp. 1325–1339, jul 2014

  20. [29]

    Spidercnn: Deep learning on point sets with parameterized convolutional filters,

    Y . Xu, T. Fan, M. Xu, L. Zeng, and Y . Qiao, “Spidercnn: Deep learning on point sets with parameterized convolutional filters,” 2018. [Online]. Available: https://arxiv.org/abs/1803.11527

  21. [30]

    Evaluating fuzz testing,

    G. Klees, A. Ruef, B. Cooper, S. Wei, and M. Hicks, “Evaluating fuzz testing,” in Proceedings of the 2018 ACM SIGSAC Conference on Computer and Communications Security , ser. CCS ’18. New York, NY , USA: Association for Computing Machinery, 2018, p. 2123–2138. [Online]. Availa...

  22. [31]

    Token-Level fuzzing,

    C. Salls, C. Jindal, J. Corina, C. Kruegel, and G. Vigna, “Token-Level fuzzing,” in 30th USENIX Security Symposium (USENIX Security 21) . USENIX Association, Aug. 2021, pp. 2795–2809. [Online]. Available: https://www.usenix.org/conference/usenixsecurity21/presentation/salls

  23. [32]

    Skyfire: Data-driven seed generation for fuzzing,

    J. Wang, B. Chen, L. Wei, and Y . Liu, “Skyfire: Data-driven seed generation for fuzzing,” in 2017 IEEE Symposium on Security and Privacy (SP), 2017, pp. 579–594

  24. [33]

    Superion: grammar-aware greybox fuzzing,

    ——, “Superion: grammar-aware greybox fuzzing,” in Proceedings of the 41st International Conference on Software Engineering , ser. ICSE ’19. IEEE Press, 2019, p. 724–735. [Online]. Available: https://doi.org/10.1109/ICSE.2019.00081

  25. [34]

    Graphfuzz: Library api fuzzing with lifetime-aware dataflow graphs,

    H. Green and T. Avgerinos, “Graphfuzz: Library api fuzzing with lifetime-aware dataflow graphs,” inProceedings of the 44th International Conference on Software Engineering , ser. ICSE ’22. New York, NY , USA: Association for Computing Machinery, 2022, p. 1070–1081. [Online]. A...

  26. [35]

    Nnsmith: Generating diverse and valid test cases for deep learning compilers,

    J. Liu, J. Lin, F. Ruffy, C. Tan, J. Li, A. Panda, and L. Zhang, “Nnsmith: Generating diverse and valid test cases for deep learning compilers,” in Proceedings of the 28th ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Vo...

  27. [36]

    Neuri: Diversifying dnn generation via inductive rule inference,

    J. Liu, J. Peng, Y . Wang, and L. Zhang, “Neuri: Diversifying dnn generation via inductive rule inference,” in Proceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering , ser. ESEC/FSE 2023. New York, NY...

  28. [37]

    Adversarial testing with rein- forcement learning: A case study on autonomous driving,

    A. Doreste, M. Biagiola, and P. Tonella, “Adversarial testing with rein- forcement learning: A case study on autonomous driving,” in 2024 IEEE Conference on Software Testing, Verification and Validation (ICST) , 2024, pp. 293–304

  29. [38]

    Towards understanding the effectiveness of large language models on directed test input generation,

    Z. Jiang, M. Wen, J. Cao, X. Shi, and H. Jin, “Towards understanding the effectiveness of large language models on directed test input generation,” in Proceedings of the 39th IEEE/ACM International Conference on Automated Software Engineering , 2024, pp. 1408–1420

  30. [2020]

    Available: https://arxiv.org/abs/2006.03997

    [Online]. Available: https://arxiv.org/abs/2006.03997

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.