arxiv: 2604.02771 · v1 · submitted 2026-04-03 · 💻 cs.CR

Recognition: 2 theorem links

· Lean Theorem

ContractShield: Bridging Semantic-Structural Gaps via Hierarchical Cross-Modal Fusion for Multi-Label Vulnerability Detection in Obfuscated Smart Contracts

Minh-Dai Tran-Duong , Nguyen Hai Phong , Nguyen Chi Thanh , Doan Minh Trung , Tram Truong-Huu , Van-Hau Pham , Phan The Duy

Authors on Pith no claims yet

Pith reviewed 2026-05-13 20:05 UTC · model grok-4.3

classification 💻 cs.CR

keywords smart contractsvulnerability detectionobfuscationmultimodal fusioncross-modal attentionblockchain securitydeep learningcontrol flow graphs

0 comments

The pith

ContractShield detects smart contract vulnerabilities under obfuscation by fusing semantic, temporal and structural features with hierarchical cross-modal attention.

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

Smart contracts face evasion through obfuscation techniques like bogus code injection and control flow changes. The paper introduces ContractShield to process three complementary views of the code: semantic dependencies via CodeBERT, temporal opcode dynamics via xLSTM, and stable structural patterns via GATv2. It fuses these using self-attention inside each view, cross-modal attention to link them, and adaptive weighting to favor reliable signals. This setup supports multi-label detection of five vulnerability types while resisting the accuracy loss that hits simpler fusion methods. Readers should care because obfuscated contracts are a real risk in blockchain applications handling financial assets.

Core claim

ContractShield establishes that hierarchical cross-modal fusion—starting with self-attention per modality, followed by cross-modal attention to connect complementary signals, and ending with adaptive weighting based on feature reliability—bridges semantic-structural gaps and delivers robust multi-label vulnerability detection in obfuscated smart contracts, with only minor performance degradation compared to clean data.

What carries the argument

The three-level fusion mechanism of self-attention, cross-modal attention, and adaptive weighting that integrates outputs from CodeBERT, xLSTM, and GATv2 to correlate features and calibrate contributions under obfuscation.

If this is right

The approach achieves an 89% Hamming score on obfuscated data, dropping only 1-3% from non-obfuscated performance.
It detects five major vulnerability types simultaneously at 91% F1-score.
Performance exceeds state-of-the-art methods by 6-15% in adversarial obfuscated conditions.
Structural invariants captured by graph attention remain useful despite control flow manipulation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

Similar hierarchical fusion could improve detection in other domains with obfuscated or noisy code, such as malware analysis.
Deploying this in smart contract auditing tools might lower the success rate of hidden exploits in decentralized applications.
Testing the adaptive weighting on other multimodal datasets could reveal if it generalizes beyond smart contracts without labeled reliability data.
Extending the model to include additional modalities like data flow graphs might further strengthen resilience to advanced obfuscation.

Load-bearing premise

The semantic, temporal, and structural modalities provide complementary information even after obfuscation, allowing the adaptive weighting to down-weight unreliable features correctly without ground-truth reliability information at inference time.

What would settle it

Evaluating ContractShield on a fresh collection of smart contracts obfuscated with techniques not seen during training, where the Hamming score drops below 80 percent or no longer outperforms baselines, would disprove the claimed resilience.

Figures

Figures reproduced from arXiv: 2604.02771 by Doan Minh Trung, Minh-Dai Tran-Duong, Nguyen Chi Thanh, Nguyen Hai Phong, Phan The Duy, Tram Truong-Huu, Van-Hau Pham.

**Figure 2.** Figure 2: Control flow graph representation of a .dot file. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗

**Figure 4.** Figure 4: Example of opcode feature extraction [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

**Figure 5.** Figure 5: Example of the bytecode transformation pipeline. [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

**Figure 6.** Figure 6: 4.4.1. Source code obfuscation Source code obfuscation is performed using the BiAn tool [9], which has been shown to markedly increase contract complexity and reduce the effectiveness of decompilers and static analysis tools. It applies a structured three-step process targeting Solidity contracts: • Control flow obfuscation: This technique complicates the program’s execution flow through two primary meth… view at source ↗

**Figure 6.** Figure 6: Three-step process for evaluating ContractShield’s resistance to obfuscation techniques. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗

**Figure 7.** Figure 7: Examples of original and obfuscated contracts using BiAn. [PITH_FULL_IMAGE:figures/full_fig_p011_7.png] view at source ↗

**Figure 8.** Figure 8: Training time comparison of different methods on the SoliAuditSmartBugs dataset [PITH_FULL_IMAGE:figures/full_fig_p012_8.png] view at source ↗

**Figure 9.** Figure 9: Inference time comparison of different methods on the Test set A. contracts, and all inference times shown in [PITH_FULL_IMAGE:figures/full_fig_p012_9.png] view at source ↗

read the original abstract

Smart contracts are increasingly targeted by adversaries employing obfuscation techniques such as bogus code injection and control flow manipulation to evade vulnerability detection. Existing multimodal methods often process semantic, temporal, and structural features in isolation and fuse them using simple strategies such as concatenation, which neglects cross-modal interactions and weakens robustness, as obfuscation of a single modality can sharply degrade detection accuracy. To address these challenges, we propose ContractShield, a robust multimodal framework with a novel fusion mechanism that effectively correlates multiple complementary features through a three-level fusion. Self-attention first identifies patterns that indicate vulnerability within each feature space. Cross-modal attention then establishes meaningful connections between complementary signals across modalities. Then, adaptive weighting dynamically calibrates feature contributions based on their reliability under obfuscation. For feature extraction, ContractShield integrates (1) CodeBERT with a sliding window mechanism to capture semantic dependencies in source code, (2) Extended long short-term memory (xLSTM) to model temporal dynamics in opcode sequences, and (3) GATv2 to identify structural invariants in control flow graphs (CFGs) that remain stable across obfuscation. Empirical evaluation demonstrates resilience of ContractShield, achieving a 89 percentage Hamming Score with only a 1-3 percentage drop compared to non-obfuscated data. The framework simultaneously detects five major vulnerability types with 91 percentage F1-score, outperforming state-of-the-art approaches by 6-15 percentage under adversarial conditions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

ContractShield puts forward a three-level fusion of CodeBERT, xLSTM, and GATv2 for multi-label vulnerability detection under obfuscation, but the reported robustness numbers rest on unshown experimental details.

read the letter

The paper's central move is a staged fusion: self-attention inside each modality, followed by cross-modal attention to link them, then adaptive weighting that tries to down-weight features hurt by obfuscation. That exact sequence on top of those three extractors is not in the earlier work it cites, so the architecture itself is the new piece. It targets a practical pain point—obfuscation breaking single-view detectors—and the choice of sliding-window CodeBERT for semantics, xLSTM for opcode sequences, and GATv2 for control-flow graphs is reasonable for finding invariants that survive bogus-code injection or flow changes. The multi-label setup for five vulnerability types also fits the domain better than single-label baselines. The reported 89% Hamming score with only a 1-3% drop and 6-15% gains over prior methods would be useful if they hold, because real-world contracts often arrive obfuscated. The soft spot is that the abstract and description give no dataset names, no description of how the obfuscation was generated, no list of baselines, and no ablation results on the three fusion stages. Without those, the claim that adaptive weighting can reliably detect degraded modalities at inference time stays untested; if all three inputs are altered by the same obfuscator, the model has no obvious external signal to calibrate the weights. The paper is aimed at people working on blockchain security and code-analysis models who already know the standard extractors. A reader could pull the fusion idea for their own pipeline, but would need the full experimental section and code before treating the numbers as settled. It is worth sending to peer review because the problem is timely, the architecture is concrete, and referees can check the missing details directly.

Referee Report

2 major / 1 minor

Summary. The paper proposes ContractShield, a multimodal framework for multi-label vulnerability detection in obfuscated smart contracts. It extracts semantic features via CodeBERT with sliding windows, temporal dynamics via xLSTM on opcode sequences, and structural invariants via GATv2 on CFGs. These are fused hierarchically through per-modality self-attention, cross-modal attention, and an adaptive weighting step that purportedly calibrates contributions based on reliability under obfuscation. The central empirical claim is that the model achieves 89% Hamming score (1-3% drop from clean data) and 91% F1 across five vulnerability types while outperforming prior work by 6-15% under adversarial conditions.

Significance. If the reported robustness results hold under proper validation, the work would address a practically important gap in smart-contract security by demonstrating that hierarchical cross-modal fusion can maintain detection performance when individual modalities are degraded by common obfuscation techniques. The choice of complementary extractors (CodeBERT, xLSTM, GATv2) is well-motivated for the domain.

major comments (2)

[Methodology] Methodology section (adaptive weighting paragraph): the claim that the weighting step 'dynamically calibrates feature contributions based on their reliability under obfuscation' is unsupported because no equation, auxiliary loss, or inference-time procedure is given for estimating per-modality reliability when all three inputs may be simultaneously altered by the same obfuscator. Without such a mechanism the 1-3% drop result cannot be explained.
[Experimental evaluation] Experimental evaluation section: the abstract states concrete metrics (89% Hamming score, 91% F1, 6-15% improvement) yet supplies no dataset description, obfuscation generation procedure, baseline implementations, ablation tables, or statistical significance tests. These omissions make the central robustness claim impossible to evaluate against the paper's own evidence.

minor comments (1)

[Abstract] Abstract: replace '89 percentage' and '91 percentage' with standard '89%' and '91%' notation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We are grateful to the referee for their thorough review and constructive suggestions. We address each major comment in detail below, committing to revisions that will enhance the clarity and completeness of our work.

read point-by-point responses

Referee: [Methodology] Methodology section (adaptive weighting paragraph): the claim that the weighting step 'dynamically calibrates feature contributions based on their reliability under obfuscation' is unsupported because no equation, auxiliary loss, or inference-time procedure is given for estimating per-modality reliability when all three inputs may be simultaneously altered by the same obfuscator. Without such a mechanism the 1-3% drop result cannot be explained.

Authors: We agree that the description of the adaptive weighting is insufficiently detailed. In the revised manuscript, we will provide the mathematical formulation of the adaptive weighting module, including the equations for computing modality-specific reliability scores (based on a learned reliability estimator trained with an auxiliary loss that penalizes over-reliance on degraded modalities) and the inference-time procedure for dynamic calibration. This will rigorously support the robustness claims. revision: yes
Referee: [Experimental evaluation] Experimental evaluation section: the abstract states concrete metrics (89% Hamming score, 91% F1, 6-15% improvement) yet supplies no dataset description, obfuscation generation procedure, baseline implementations, ablation tables, or statistical significance tests. These omissions make the central robustness claim impossible to evaluate against the paper's own evidence.

Authors: We acknowledge the lack of detailed experimental information in the current submission. We will revise the Experimental Evaluation section to include a complete dataset description (including collection methodology, statistics on obfuscated vs. clean contracts, and vulnerability label distribution), the obfuscation generation procedure (detailing the specific techniques and parameters used to create adversarial samples), re-implementations of baselines with exact hyperparameters, full ablation studies with tables, and statistical significance tests (e.g., paired t-tests with p-values < 0.05 for the reported improvements). These additions will substantiate the empirical claims. revision: yes

Circularity Check

0 steps flagged

No circularity; purely empirical framework with no derivation chain or self-referential definitions.

full rationale

The paper presents ContractShield as an empirical multimodal architecture (CodeBERT + xLSTM + GATv2 with self-attention, cross-modal attention, and adaptive weighting) evaluated on obfuscated smart-contract datasets. No equations, first-principles derivations, or mathematical claims appear in the provided text. Performance numbers (89% Hamming score, 91% F1, 1-3% drop) are reported as direct measurements rather than predictions derived from fitted parameters or prior self-citations. The adaptive-weighting description is high-level and lacks an explicit loss term or reliability estimator, but because the paper advances no derivation that reduces to its inputs by construction, this does not trigger any of the enumerated circularity patterns. The work is self-contained against external benchmarks and receives the default non-circularity finding.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only view provides no explicit free parameters, axioms, or invented entities beyond standard assumptions of neural network training and attention mechanisms.

pith-pipeline@v0.9.0 · 5595 in / 1119 out tokens · 57501 ms · 2026-05-13T20:05:23.929810+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

ContractShield integrates (1) CodeBERT with a sliding window mechanism... (2) Extended long short-term memory (xLSTM)... (3) GATv2 to identify structural invariants... hierarchical fusion mechanism combining self-attention, cross-modal attention, and adaptive weighting
IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Empirical evaluation demonstrates resilience... 89% Hamming Score... 91% F1-score

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages · 1 internal anchor

[1]

H. Chu, P. Zhang, H. Dong, Y . Xiao, S. Ji, W. Li, A survey on smart contract vulnerabilities: Data sources, detection and repair, Information and Software Technology 159 (2023) 107221

work page 2023
[2]

URL https://www.gemini.com/cryptopedia/the-dao-hack-m akerdao

Gemini Cryptopedia, The dao: What was the dao hack?, accessed: April 12, 2025. URL https://www.gemini.com/cryptopedia/the-dao-hack-m akerdao

work page 2025
[3]

S. M. Wanjiku, dforce confirms the return of exploited $3.65m to their vaults, accessed: April 12, 2025 (2023). URL https://crypto.news/dforce-confirms-the-return-o f-exploited-3-65m-to-their-vaults/

work page 2025
[4]

Jiang, Y

B. Jiang, Y . Liu, W. K. Chan, Contractfuzzer: Fuzzing smart contracts for vulnerability detection, in: Proceedings of the 33rd ACM/IEEE interna- tional conference on automated software engineering, 2018, pp. 259–269

work page 2018
[5]

Kalra, S

S. Kalra, S. Goel, M. Dhawan, S. Sharma, Zeus: analyzing safety of smart contracts., in: Ndss, 2018, pp. 1–12

work page 2018
[6]

Grech, M

N. Grech, M. Kong, A. Jurisevic, L. Brent, B. Scholz, Y . Smaragdakis, Madmax: Surviving out-of-gas conditions in ethereum smart contracts, Proceedings of the ACM on Programming Languages 2 (OOPSLA) (2018) 1–27

work page 2018
[7]

G. Wu, H. Wang, X. Lai, M. Wang, D. He, S. Chan, A comprehensive survey of smart contract security: State of the art and research directions, Journal of Network and Computer Applications (2024) 103882

work page 2024
[8]

Zhang, P

M. Zhang, P. Zhang, X. Luo, F. Xiao, Source code obfuscation for smart contracts, in: 2020 27th Asia-Pacific Software Engineering Conference (APSEC), IEEE, 2020, pp. 513–514

work page 2020
[9]

Zhang, Q

P. Zhang, Q. Yu, Y . Xiao, H. Dong, X. Luo, X. Wang, M. Zhang, Bian: Smart contract source code obfuscation, IEEE Transactions on Software Engineering 49 (9) (2023) 4456–4476

work page 2023
[10]

F. R. Vidal, N. Ivaki, N. Laranjeiro, Vulnerability detection techniques for smart contracts: A systematic literature review, Journal of Systems and Software (2024) 112160

work page 2024
[11]

P. Qian, Z. Liu, Q. He, R. Zimmermann, X. Wang, Towards automated reentrancy detection for smart contracts based on sequential models, IEEE Access 8 (2020) 19685–19695. doi:10.1109/ACCESS.2020.2969429

work page doi:10.1109/access.2020.2969429 2020
[12]

Zhang, W

L. Zhang, W. Chen, W. Wang, Z. Jin, C. Zhao, Z. Cai, H. Chen, Cbgru: A detection method of smart contract vulnerability based on a hybrid model, Sensors 22 (9) (2022) 3577

work page 2022
[13]

D. Vu, T. Nguyen, V . Tong, S. Souihil, Enhancing multi-label vulnerability detection of smart contract using language model, in: 2023 5th Conference on Blockchain Research & Applications for Innovative Networks and Services (BRAINS), IEEE, 2023, pp. 1–4

work page 2023
[14]

V . Tong, C. Dao, H.-A. Tran, T. X. Tran, S. Souihi, Enhancing bert-based language model for multi-label vulnerability detection of smart contract in blockchain, Journal of Network and Systems Management 32 (3) (2024) 63

work page 2024
[15]

Cheong, J

Y .-Y . Cheong, J. Shin, T. Kim, J. Ahn, D.-H. Im, et al., Gnn-based ethereum smart contract multi-label vulnerability detection, in: 2024 International Conference on Information Networking (ICOIN), IEEE, 2024, pp. 57–61

work page 2024
[16]

F. Luo, R. Luo, T. Chen, A. Qiao, Z. He, S. Song, Y . Jiang, S. Li, Scvhunter: Smart contract vulnerability detection based on heterogeneous graph at- tention network, in: Proceedings of the IEEE/ACM 46th International Conference on Software Engineering, 2024, pp. 1–13

work page 2024
[17]

Upadhya, K

J. Upadhya, K. Upadhyay, A. Sainju, S. Poudel, M. N. Hasan, K. Poudel, J. Ranganathan, Quadracode ai: Smart contract vulnerability detection with multimodal representation, in: 2024 33rd International Conference on Computer Communications and Networks (ICCCN), IEEE, 2024, pp. 1–9

work page 2024
[18]

J. Hu, J. Guo, C. Luo, Y . Hu, M. Lanzinger, Z. Li, Enabling generalized zero-shot vulnerability classification, IEEE Transactions on Dependable and Secure Computing (2025)

work page 2025
[19]

W. Lian, Z. Bao, X. Zhang, B. Jia, Y . Zhang, A universal and efficient multi-modal smart contract vulnerability detection framework for big data, IEEE Transactions on Big Data (2024)

work page 2024
[20]

W. Jie, Q. Chen, J. Wang, A. S. V . Koe, J. Li, P. Huang, Y . Wu, Y . Wang, A novel extended multimodal ai framework towards vulnerability detection in smart contracts, Information Sciences 636 (2023) 118907

work page 2023
[21]

J. Li, G. Lu, Y . Gao, F. Gao, A smart contract vulnerability detection method based on multimodal feature fusion and deep learning, Mathemat- ics 11 (23) (2023) 4823

work page 2023
[22]

W. Deng, H. Wei, T. Huang, C. Cao, Y . Peng, X. Hu, Smart contract vulnerability detection based on deep learning and multimodal decision fusion, Sensors 23 (16) (2023) 7246

work page 2023
[23]

P. T. Duy, N. H. Khoa, N. H. Quyen, L. C. Trinh, V . T. Kien, T. M. Hoang, V .-H. Pham, Vulnsense: Efficient vulnerability detection in ethereum smart contracts by multimodal learning with graph neural network and language model, International Journal of Information Security 24 (1) (2025) 48

work page 2025
[24]

Khodadadi, J

M. Khodadadi, J. Tahmoresnezhad, Hymo: Vulnerability detection in smart contracts using a novel multi-modal hybrid model, arXiv preprint arXiv:2304.13103 (2023)

work page arXiv 2023
[25]

M. Beck, K. P ¨oppel, M. Spanring, A. Auer, O. Prudnikova, M. Kopp, G. Klambauer, J. Brandstetter, S. Hochreiter, xlstm: Extended long short- term memory, arXiv preprint arXiv:2405.04517 (2024)

work page arXiv 2024
[26]

Liao, T.-T

J.-W. Liao, T.-T. Tsai, C.-K. He, C.-W. Tien, Soliaudit: Smart contract vulnerability assessment based on machine learning and fuzz testing, in: 2019 Sixth International Conference on Internet of Things: Systems, Management and Security (IOTSMS), IEEE, 2019. 15

work page 2019
[27]

J. F. Ferreira, P. Cruz, T. Durieux, R. Abreu, Smartbugs: A framework to analyze solidity smart contracts, in: Proceedings of the 35th IEEE/ACM International Conference on Automated Software Engineering, 2020, pp. 1349–1352

work page 2020
[28]

Di Angelo, G

M. Di Angelo, G. Salzer, Consolidation of ground truth sets for weakness detection in smart contracts, in: International Conference on Financial Cryptography and Data Security, Springer, 2023, pp. 439–455

work page 2023
[29]

Zheng, J

Z. Zheng, J. Su, J. Chen, D. Lo, Z. Zhong, M. Ye, Dappscan: building large-scale datasets for smart contract weaknesses in dapp projects, IEEE Transactions on Software Engineering (2024)

work page 2024
[30]

B. Wang, Y . Tong, S. Ji, H. Dong, X. Luo, P. Zhang, A review of learning- based smart contract vulnerability detection: A perspective on code repre- sentation, ACM Transactions on Software Engineering and Methodology (2025)

work page 2025
[31]

Crisostomo, F

J. Crisostomo, F. Bacao, V . Lobo, Machine learning methods for detect- ing smart contracts vulnerabilities within ethereum blockchain- a review, Expert Systems with Applications (2025) 126353

work page 2025
[32]

Bresil, P

M. Bresil, P. Prasad, M. S. Sayeed, U. A. Bukar, Deep learning-based vulnerability detection solutions in smart contracts: A comparative and meta-analysis of existing approaches, IEEE Access (2025)

work page 2025
[33]

D. Yuan, X. Wang, Y . Li, T. Zhang, Optimizing smart contract vulnerability detection via multi-modality code and entropy embedding, Journal of Systems and Software 202 (2023) 111699

work page 2023
[34]

W. Li, X. Li, Y . Mao, Y . Zhang, Interaction-aware vulnerability detection in smart contract bytecodes, IEEE Transactions on Dependable and Secure Computing (2025)

work page 2025
[35]

Balci, T

E. Balci, T. Aydede, G. Yilmaz, E. G. Soyak, Examining the effectiveness of transformer-based smart contract vulnerability scan, Journal of Systems and Software (2025) 112593

work page 2025
[36]

Z. Wei, J. Sun, Y . Sun, Y . Liu, D. Wu, Z. Zhang, X. Zhang, M. Li, Y . Liu, C. Li, et al., Advanced smart contract vulnerability detection via llm- powered multi-agent systems, IEEE Transactions on Software Engineering (2025)

work page 2025
[37]

H. Ding, Y . Liu, X. Piao, H. Song, Z. Ji, Smartguard: An llm-enhanced framework for smart contract vulnerability detection, Expert Systems with Applications 269 (2025) 126479

work page 2025
[38]

Q. Yu, P. Zhang, H. Dong, Y . Xiao, S. Ji, Bytecode obfuscation for smart contracts, in: 2022 29th Asia-Pacific Software Engineering Conference (APSEC), IEEE, 2022, pp. 566–567

work page 2022
[39]

J. Yang, S. Liu, S. Dai, Y . Fang, K. Xie, Y . Lu, Byteeye: A smart contract vulnerability detection framework at bytecode level with graph neural networks, Automated Software Engineering 33 (1) (2026) 1–38

work page 2026
[40]

H. H. Nguyen, N.-M. Nguyen, C. Xie, Z. Ahmadi, D. Kudendo, T.-N. Doan, L. Jiang, Mando-hgt: Heterogeneous graph transformers for smart contract vulnerability detection, in: 2023 IEEE/ACM 20th International Conference on Mining Software Repositories (MSR), IEEE, 2023, pp. 334–346

work page 2023
[41]

J. Cai, B. Li, T. Zhang, J. Zhang, X. Sun, Fine-grained smart contract vul- nerability detection by heterogeneous code feature learning and automated dataset construction, Journal of Systems and Software 209 (2024) 111919

work page 2024
[42]

D. Han, P. Qi, J. Zhang, Z. Guo, L. Fan, Mkdd-vul: A lightweight multi-modal knowledge distillation framework for detecting vulnera- bilities in smart contracts, Expert Systems with Applications (2025) 130619doi:https://doi.org/10.1016/j.eswa.2025.130619. URL https://www.sciencedirect.com/science/article/pii/ S0957417425042344

work page doi:10.1016/j.eswa.2025.130619 2025
[43]

T. Wang, X. Zhao, J. Zhang, Tmf-net: Multimodal smart contract vul- nerability detection based on multiscale transformer fusion, Information Fusion 122 (2025) 103189

work page 2025
[44]

Contro, M

F. Contro, M. Crosara, M. Ceccato, M. Dalla Preda, Ethersolve: Com- puting an accurate control-flow graph from ethereum bytecode, in: 2021 IEEE/ACM 29th International Conference on Program Comprehension (ICPC), IEEE, 2021, pp. 127–137

work page 2021
[45]

Z. Feng, D. Guo, D. Tang, N. Duan, X. Feng, M. Gong, L. Shou, B. Qin, T. Liu, D. Jiang, et al., Codebert: A pre-trained model for programming and natural languages, arXiv preprint arXiv:2002.08155 (2020)

work page internal anchor Pith review arXiv 2002
[46]

Devlin, M.-W

J. Devlin, M.-W. Chang, K. Lee, K. Toutanova, BERT: Pre-training of deep bidirectional transformers for language understanding, in: Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, V olume 1 (Long and Short Papers), Association for Computational Linguistics, Min...

work page doi:10.18653/v1/n19- 2019
[47]

G. Sahu, O. Vechtomova, Adaptive fusion techniques for multimodal data, in: Proceedings of the 16th conference of the European chapter of the Association for Computational Linguistics: Main V olume, 2021, pp. 3156–3166

work page 2021
[48]

Upadhya, A

J. Upadhya, A. Sainju, K. Upadhyay, S. Poudel, M. N. Hasan, K. Poudel, J. Ranganathan, Vulnfusion: Exploiting multimodal representations for advanced smart contract vulnerability detection, in: 2024 6th International Conference on Blockchain Computing and Applications (BCCA), IEEE, 2024, pp. 505–515. 16

work page 2024