We introduce a new polynomial-time graph invariant combining three complementary components: (1) the enriched Bipartite Double Cover (BDC) vertex cover signature; (2) the Laplacian eigenvalue spectrum; and (3) the K_4 clique profile and Ollivier-Ricci curvature. We demonstrate that while spectral and standard BDC methods collapse on strongly regular graphs (SRGs) due to extreme symmetry, the geometric and dense-topological components of this hybrid invariant break the cospectrality. Specifically, the invariant successfully discriminates the classical cospectral pair Shrikhande vs. Rook(4,4) [SRG(16,6,2,2)] in polynomial time, where Shrikhande is K_4-free and Rook(4,4) contains exactly 8 cliques. Furthermore, empirical validation on low-power ARM Edge hardware demonstrates that the invariant scales efficiently, processing dense Paley graphs (up to N=97) in under 5 milliseconds. This confirms its sub-millisecond viability for real-time edge computing, zero-knowledge proofs (ZKPs) cryptanalysis, and cheminformatics.
Zhuoran Pan, Yue Li (102191), Zhi Guan, Jianbin Hu · 5 authors
The emergence of Large Language Models (LLMs) offers a transformative interface for Web3, yet existing benchmarks fail to capture the complexity of translating high-level user intents into functionally correct, state-dependent on-chain transactions. We present \textsc{Intent2Tx}, a high-fidelity benchmark featuring 29,921 single-step and 1,575 multi-step instances meticulously derived from 300 days of real-world Ethereum mainnet traces. Unlike prior works that rely on synthetic instructions, \textsc{Intent2Tx} grounds natural language intents in real-world protocol interactions across 11 categories, including diverse long-tail Decentralized Finance (DeFi) primitives. To enable rigorous evaluation, we propose an execution-aware framework that transcends surface-level text matching by employing differential state analysis on forked mainnet environments. Our extensive evaluation of 16 state-of-the-art LLMs reveals that while scaling and retrieval-augmentation enhance logical consistency and parameter precision, current models struggle with out-of-distribution generalization and multi-step planning. Crucially, our execution-based analysis demonstrates that syntactically valid outputs often fail to achieve intended state transitions, highlighting a significant gap in current "reasoning-to-execution" capabilities. \textsc{Intent2Tx} serves as a critical foundation for developing autonomous, reliable agents in intent-centric Web3 ecosystems. Code and data: https://anonymous.4open.science/r/Intent2Tx_Bench-97FF .
Security vulnerabilities in smart contracts can lead to serious economic losses, while traditional detection methods have limitations in coarse-grained labeling and single graph structure representation. We propose an improved GraphCodeBERT based multi granularity and multi graph fusion smart contract vulnerability detection method. Firstly, propose a vulnerability multi granularity annotation (MA) module that includes file level, function level, and statement level to address the issue of fine-grained annotation. Secondly, based on the data flow graph encoded by GraphCodeBERT, further design a multi graph fusion representation learning (MGRL) strategy. This strategy integrates control flow graph and function call graph, captures code semantics from control dependency and call dependency dimensions through graph attention network and graph convolutional network, to comprehensively enhance the representation capability of complex vulnerabilities. The detection results of four smart contract vulnerabilities show that our method outperforms mainstream methods such as Slither, CodeBERT, Devign, and GraphCodeBERT in accuracy, recall, and F1 score. Especially in sentence level detection tasks, the F1 score reached 81.03%, an increase of 7.31% compared to the baseline, and the false positive rate decreased to 8.15%.
The transition from static articles to a living Scientific Knowledge Graph, as proposed in our previous work, promises to accelerate discovery and restore feedback loops in science. However, a fully open, semantically linked graph of all scientific knowledge also presents an unprecedented dual-use risk: it could become a roadmap for malicious actors to identify and exploit hidden vulnerabilities. This paper addresses that paradox by introducing a comprehensive framework for "Ethical Immunity" — a set of proactive, architecture-level mechanisms designed to make the Knowledge Graph resilient to misuse without resorting to censorship or secrecy. We detail a three-pillar system: (1) AI-powered Red and Blue Teams that continuously simulate misuse scenarios and generate countermeasures; (2) Decentralized Autonomous Organizations (DAOs) for transparent, expert-driven oversight and risk assessment; and (3) "Ethical Quarantine" protocols that allow for the temporary isolation of high-risk knowledge while ensuring the parallel development of defenses. We argue that such a framework transforms the Knowledge Graph from a passive repository into an active immune system for civilization, capable of identifying and neutralizing threats at the speed of discovery. This paper provides a technical and organizational blueprint for building safety into the very fabric of 21st-century science.
Multi-agent architectures leveraging Large Language Models (LLMs) have significantly advanced the precision of Question Answering (QA) systems across diverse domains. However, existing frameworks remain vulnerable to adversarial manipulations, including poisoning, backdoor, and jailbreak at tacks, primarily due to their reliance on centralized orchestration. To mitigate these risks, we propose AgentChain, a framework that substitutes centralized control with a distributed semantic consensus process. By modeling the blockchain as an ideal functionality, AgentChain establishes a secure distributed layer to coordinate role allocation, answer proposal, evaluation and voting through a decentralized council. Specifically, we design Proof-of-Content-Quality (PoCQ) mechanism to ensure that the f inal answers reflect a robust semantic agreement among the majority of honest agents. Furthermore, we propose an incentive mechanism based on stake reassignment that penalizes malicious agents by reducing their rewards, ultimately phasing them out of the network. Comprehensive evaluations across eight datasets demonstrate that AgentChain achieves superior performance and resilience. AgentChain minimizes the impact of poisoning attacks on precision to less than 3% and reduces the success rate of backdoor and jailbreak attacks to less than 4%. These findings highlight the effectiveness and trustworthiness of AgentChain in mitigating security threats while maintaining high QA accuracy.
Fraudulent activities on blockchain networks threaten the integrity and reliability of decentralized finance ecosystems. Accurately identifying malicious nodes such as phishing or ransomware addresses, within large-scale blockchain transaction graphs remains a critical challenge due to their dynamic, sparse, and continuously evolving topologies. Transfer learning offers a powerful paradigm for fraud detection because many fraudulent schemes, including ransomware and phishing, are often orchestrated by overlapping actor groups that share behavioral and structural patterns across networks. Leveraging these shared representations enables knowledge transfer from previously observed fraud types to emerging ones. However, the complex and multi-modal nature of digital financial systems introduces substantial challenges for graph-based transfer learning. Fraudulent activities are shaped by diverse modalities including graph structure, transaction sequences, temporal price dynamics, and textual metadata, while distributional shifts frequently occur across time and platforms. Existing graph transfer learning methods struggle to model such multi-modal dependencies and to align divergent feature distributions. To tackle these challenges, we develop a Multi-mOdal Enhanced Graph Transfer Learning (MOE-GTL) framework which incorporates graph, temporal, and textual modalities for fraudulent node detection. We further introduce Temporal-aware Maximum Mean Discrepancy (TMMD), a regularization mechanism that explicitly aligns multi-modal feature distributions between source and target graphs over time. Extensive experiments reveal that our MOE-GTL model notably improves the accuracy of fraudulent node classifications on Ethereum and Solana transaction graphs.
Against the complex characteristics of the Ethereum transaction network and the limitations of existing graph embedding methods based on random walks, which fail to effectively capture transaction temporal dynamics and the flow of funds, we propose a fraud detection algorithm for Ethereum, ETX2Vec (Ethereum Transactions (TX) to Vector), which improves upon transaction subgraph construction and random walk strategies. First, in terms of transaction subgraph construction, we extract the first-order predecessor and successor neighboring nodes of the target node to reconstruct the transaction subgraph, enabling the random walk to effectively capture the complete flow of funds. Second, in the design of the random walk strategy, we introduce two key improvements: (1) the next node is selected based on the non-decreasing principle of transaction timestamps, effectively capturing the temporal dynamics of transactions within the network, and (2) a biased random walk strategy is designed based on both transaction timestamps and amounts, with a parameter α introduced to control the weighting of these factors when calculating transition probabilities. Experimental results show that ETX2Vec achieves an average performance of 96.04% in downstream node classification tasks, outperforming the best model in similar studies by 3.74%, and even surpassing neural network models such as GAT and GCN. This demonstrates that ETX2Vec is more effective at understanding and processing the Ethereum transaction network, leading to the learning of high-quality node embedding vectors.
Federated graphs learning for graphs enables multiple clients to share model knowledge and engage in collaborative training while ensuring user data privacy. Nevertheless, federated learning for graphs also faces various security threats, such as privacy leakage and malicious attacks. On the other hand, compared with other security strategies, differential privacy offers low cost and high efficiency in protecting data in federated learning, yet it can compromise the training accuracy of federated learning for graphs and, in some cases, severely degrade training performance. Therefore, this paper considers noise-sensitive scenarios where even a small amount of noise can significantly impact training, and integrates knowledge distillation with distributed differential privacy federated learning for graphs. This approach enhances model training accuracy under noise-sensitive conditions while mitigating the adverse effects of differential privacy noise on training, all while ensuring model security. In addition to leveraging differential privacy to protect data and parameter privacy, we further aim to defend against malicious client attacks. By establishing a global consensus on the gradient clipping range, we use zero-knowledge proofs to provide sampled verification of the gradient range, demonstrating that the parameters uploaded by clients have been correctly clipped during training. Parameters that fail the verification are discarded, thereby further enhancing security.
Privacy-Preserving Technologies in Data
Advanced Graph Neural Networks
Distributed Sensor Networks and Detection Algorithms
Large language models and retrieval-augmented generation systems treat all knowledge as uniformly persistent, ignoring a well-established property of information: that different types of knowledge expire at fundamentally different rates. This paper introduces the Dynamic Epistemic Decay Framework, a formal multi-dimensional theory that characterizes knowledge validity as a function of five independent decay dimensions: temporal decay (𝜆𝜆𝑡𝑡), paradigm decay (𝜆𝜆𝑝𝑝), uncertainty decay (𝜆𝜆𝑢𝑢), dependency decay (𝜆𝜆𝑑𝑑), and zero decay (𝜆𝜆0). We implement this framework as a four-phase retrieval pipeline and evaluate it on the TempQuestions benchmark (n=1,740) against three baselines: standard cosine similarity, BM25 lexical retrieval, and naive recency ranking. Decay-weighted retrieval achieves 92.1% accuracy versus 13.5% for standard semantic retrieval—a 78.6 percentage point improvement—with zero regressions on stable factual queries. On semantically complex temporal benchmarks where lexical heuristics fail, the framework dominates a more resourced BM25 baseline (90.2% vs 1.6% on date-bounded role queries). Epistemic modulation (Phase 4) and dependency graph reasoning (Phase 3) further demonstrate correct mechanism behavior on specialized benchmarks, validated via proof-of-concept implementation. Unlike temporal KG completion approaches that require structured annotation, and unlike contrastive training approaches to time-sensitive RAG, the decay framework is training-free and operates directly over unstructured text corpora. We argue that the decay framework completes the separation of concerns that RAG began: decoupling not just factual storage from model parameters, but factual currency from both.
Pre-trained language models (PLMs) have shown strong potential in Ethereum account modeling and fraud detection. However, existing approaches often overlook the graph-structured nature of transaction networks. In addition, they struggle with the long-tail distribution of account activity, resulting in anisotropic embedding spaces and poor representation quality for low-frequency accounts. In this paper, we present IGT4ETH, a pre-trained Graph Transformer with an isotropy-enhanced post-processing, which explicitly models transaction topologies and mitigates representational anisotropy for Ethereum account classification. IGT4ETH improves structural representation by incorporating structural centrality and role embeddings into an Edge-augmented Graph Transformer, effectively capturing both topological and interaction patterns in transaction graphs. To further mitigate embedding anisotropy, we systematically evaluate various post-processing techniques. Among them, we adopt the Conceptor Negation (CN) method to softly suppress latent features dominated by high-frequency words via matrix conceptors, alongside a modified Focal-InfoNCE loss to enhance directional uniformity and representation balance. Extensive experiments on four real-world Ethereum account classification tasks, including phishing, exchange, mining, and ICO-wallet classification, demonstrate that IGT4ETH consistently outperforms state-of-the-art PLM-based baselines in terms of classification performance.
Open access
Advanced Graph Neural Networks
Topic Modeling
Artificial Intelligence in Healthcare and Education
Zhaohuang Chen, Zhongqi Fu, Tao Liang, Haidong Ma · 6 authors
Abstract Since the proposal of the blockchain, its application scenarios have been continuously expanded. However, the anonymity feature of the blockchain has hindered market regulation, leading to numerous illegal activities such as phishing fraud, which has now become a serious type of crime. Currently, most phishing fraud detection technologies on blockchain platforms use transaction data to construct basic raw transaction graphs and then use neural network methods to mine key information. This study proposes a graph gated recurrent neural network (GGRNN) model that fully integrates temporal and spatial information, effectively utilizing time-related information in the transaction graph. It first takes an account as the center node to obtain its second-order transaction data and then constructs a dynamic transaction graph (DTG). Subsequently, the DTG is fed to the GGRNN to process the temporal features in a gated recurrent unit (GRU) framework and introduce graph convolutional network (GCN) operations to fully use the node neigh-bourhood topology features, obtain the embedded representation of the graph, and then perform graph classification for phishing node detection. To verify the effectiveness of the proposed model, it was applied to real-world Ethereum transaction datasets. Numerical results show that the proposed GGRNN model significantly outperforms state-of-the-art methods.
We present Viturka, a blockchain architecture that replaces wasteful proof-of-work mining with productive federated learning. The core innovation is Proof of Credibility (PoC): a consensus mechanism where block production probability is determined by accumulated reputation from validated AI contributions rather than computational hash power or financial stake. Viturka leverages recent breakthroughs in Zero-Knowledge Machine Learning (ZKML) to achieve cryptographic verification of model training. Validators generate zero-knowledge proofs attesting to correct training execution, enabling instant on-chain verification without trusted intermediaries or statistical consensus mechanisms. By integrating frameworks like EZKL and Lagrange's DeepProve with GPU-accelerated proving via the Icicle library, validation that previously required hours of recomputation now produces mathematical proofs verifiable in milliseconds. Participants earn credibility by contributing quality training data or validating others' contributions. Only the top 10 highest-credibility validators can participate in validation rounds, with mandatory cooldown periods ensuring rotation. The system uses a temporal commit-reveal scheme for data contributions combined with ZK proofs for validation—fake contributions result in permanent bans, while fraudulent validation is mathematically impossible. This creates infrastructure for training AI models on distributed data without central coordination, with economic incentives aligned toward data quality rather than raw computation. Applications range from commercially valuable use cases like DeFi credit scoring—which could unlock over $100B in overcollateralized capital—to public-good AI for rare diseases, minority languages, and environmental monitoring.
Bikki Kumar, Dev Karan, Adarsh Kandu, Ashish Khari
The social impact of decentralised online communities, such as blockchain-based social networks, is complex because their decentralisation allows users to exercise greater freedom and independence. A novel Dynamic Graph Neural Network with Temporal Knowledge Distillation (DGNN-TKD) is proposed to model and predict influence patterns. DGNN-TKD differs from standard Graph Neural Networks (GNNs), which function under the assumption of static graphs. It tracks the temporal evolution of a graph, and introduces a knowledge distillation mechanism that enables the transfer of influence embeddings over time. We propose a novel multi-dimensional influence metric that captures agent reputation, engagement and trust, supplemented with robust attention-based temporal aggregation. In experiments on decentralized social network datasets, DGNN-TKD surpasses current dynamic GNNs in influence prediction, community detection, and misinformation detection in decentralized governance/Web3 applications. This framework connects graph-based learning and social dynamics and serves as a powerful tool to study decentralized phenomena.
Abstract: This paper introduces Knowledge Tensor Lock (KTL), a novel cognitive-structural authentication framework. Unlike conventional mechanisms (passwords, biometrics), KTL anchors identity in the topology of a user’s private semantic associative network. We formalize cognition as a high-rank tensor and verify identity through an interactive challenge-response reconstruction of subgraph structures. Key Contributions: Formalization of the Knowledge Tensor ($\mathcal{K}$) and its graph projection ($G$). Introduction of the Spectral Sketch ($\mathcal{SS}$) for privacy-preserving structural storage. Analysis of heuristic security against AI-adaptive adversaries and model extraction. A roadmap for integrating Zero-Knowledge Proofs (ZKP) for decentralized identity. Note: This is a stabilized preprint (v1.2) intended for establishing conceptual priority in the fields of AI security and cognitive cryptography.
Background Cross-domain federated learning is an innovative machine learning paradigm that allows data owners from different domains to collaboratively train a shared model while preserving data privacy. However, cross-domain federated learning also faces numerous challenges, such as data and system heterogeneity, client reputation management, and potential threats from malicious attackers. Methods To address these issues, this article proposes a secure cross-domain federated learning scheme based on blockchain fair payment. The proposed scheme effectively evaluates and updates the reputation of each client through a reputation management mechanism and allocates fair rewards based on their contributions. Additionally, the scheme employs advanced cryptographic technologies such as blockchain and zero-knowledge proofs to ensure the security and fairness of data and transactions. A series of experiments are conducted to evaluate the performance and fairness of the proposed scheme on multiple datasets and models, and comparisons are conducted with other mainstream federated learning algorithms. MNIST Dataset is available at: https://www.kaggle.com/datasets/hojjatk/mnist-dataset . Fashion-MNIST Dataset is available at https://github.com/zalandoresearch/fashion-mnist . CIFAR-10 Dataset is available at https://www.cs.toronto.edu/~kriz/cifar.html . Results The experimental results demonstrate that the proposed scheme ensures the performance of federated learning while also maintaining its fairness and security. Specifically, the method achieves a test accuracy of 97% on the MNIST dataset, outperforming Federated Averaging (FedAvg) (95%) and Stochastic Controlled Averaging for Federated Learning (SCAFFOLD) (96%). On the FEMNIST dataset, it attains 89% accuracy. In terms of convergence speed, the proposed optimization-based reputation method converges in 26 rounds, which is faster than baseline methods (28–32 rounds). Under data tampering attacks (50-client scenario), the accuracy drop is less than 3%, showing strong robustness. For fairness, the trust difference and reward difference are reduced to 0.10 and 0.08, respectively. The proposed scheme significantly improves the accuracy, convergence speed, robustness, and fairness of cross-domain federated learning, advancing its practical deployment in real-world scenarios. The experimental data is available at: https://zenodo.org/records/15210778 .
Changhao Wu, Luyu Chen, Kai Wang, Weili Han · 5 authors
Occurring approximately once or twice in each block, sandwich attacks threaten Ethereum’s ecosystem by manipulating prices through strategically placed buy and sell transactions around pending user trades. Existing detection methods primarily rely on rigid heuristic rules, limiting their ability to detect increasingly sophisticated and dynamic attack variants, particularly those residing only in the mempool or spanning multiple blocks. In this paper, we propose SandWatch , a novel Ethereum sandwich attack detection framework that integrates a dual-task graph neural network (Dual-GNN) with heuristic methods. The framework comprises three main components. (1) An order-independent heuristic that captures fundamental token transfer patterns, reducing transaction volume by over 94% for subsequent graph analysis. (2) A Dual-GNN that simultaneously classifies sandwich attack transactions and DEX nodes, dynamically updating an address label pool to enhance accuracy and generalizability. (3) A positive-unlabeled learning strategy to leverage large-scale unlabeled data effectively. We first evaluate Dual-GNN on a benchmark dataset derived from publicly available Ethereum sandwich attack data, achieving an F1-score of 99.78%, outperforming the single-task baseline by 0.93 percentage points. We then deploy SandWatch on Ethereum transactions collected through blockchain interfaces and mempool pre-execution from January to May 2024. SandWatch detects 563,453 sandwich attacks, including 24,404 multi-attack, 4,902 cross-block, and 1,057 pool-failure variants, achieving an overall recall of 98.63% compared to the state-of-the-art benchmark platform. These results demonstrate the robustness of SandWatch in detecting sophisticated sandwich attacks within the real-world Ethereum ecosystem.
NFT prices are shaped by heterogeneous signals including visual appearance, textual narratives, transaction trajectories, and on-chain interactions, yet existing studies often model these factors in isolation and rarely unify multimodal alignment, temporal non-stationarity, and heterogeneous relational dependencies in a leakage-safe forecasting setting. We propose MM-Temporal-Graph, a cross-modal temporal graph transformer framework for explainable NFT valuation and information-centric risk forecasting. The model encodes image, text, transaction time series, and blockchain behavioral features, constructs a heterogeneous NFT interaction graph (co-transaction, shared creator, wallet relation, and price co-movement), and jointly performs relation-aware graph attention and global temporal–structural transformer reasoning with an adaptive fusion gate. A contrastive multimodal alignment objective improves robustness under market drift, while a risk-aware regularizer and a multi-source risk index enable early warning and interpretable attribution across modalities, time segments, and relational neighborhoods. On MultiNFT-T, MM-Temporal-Graph improves MAE from 0.162 to 0.153 and R2 from 0.823 to 0.841 over the strongest multimodal graph baseline, and achieves 87.4% early risk detection accuracy. These results support accurate, robust, and explainable NFT valuation and proactive risk monitoring in Web3 markets.
With the widespread use of smart contracts in blockchain applications, particularly in the consumer Internet of Things (CIoT), the security of smart contracts has become increasingly critical. CIoT refers to an IoT environment where various consumer devices are intelligently connected and interact via the internet. In this context, smart contracts are frequently used to automate tasks such as device control, data sharing, and transaction verification. However, vulnerabilities in smart contracts can lead to system attacks, thereby compromising the security of the entire IoT network. The Cascade Graph Convolutional Network-based Vulnerability Detection Framework for smart contracts (CGCN-DF) proposed in this study effectively enhances the security assessment of smart contracts in CIoT applications. Devices and systems in CIoT environments typically exhibit high heterogeneity and complex interaction patterns, but a single graph structure cannot fully capture the multidimensional behavioral features and complex structural relationships of smart contracts. To address this issue, this paper introduces the Semantic Contract Graph (SCG), which integrates three graph representations—Abstract Syntax Tree (AST), Control Flow Graph (CFG), and Data Flow Graph (DFG)—into a unified graph structure, comprehensively covering different aspects of the code. Furthermore, as smart contracts in CIoT environments often involve real-time data flows and complex execution paths, the CGCN-DF framework employs a cascading mechanism that performs three-level graph convolutional processing through relational, Meta-path structures, and cyclic structures. This approach extracts rich and complementary information from multi-level features, explicitly models the dynamic interactions among syntax, execution paths, and data flows, and enhances the model’s contextual awareness of vulnerability-triggering conditions. Ultimately, the framework achieves coarse-grained detection at the contract level and fine-grained detection at the line level. Experimental results demonstrate that the proposed method can effectively localize vulnerabilities down to specific code lines. This not only enhances the precision and practicality of smart contract vulnerability detection but, more importantly, contributes an innovative technical framework and theoretical methodology to the CIoT field—particularly in areas such as DSL program analysis and the intersection of IoT and blockchain security. This work thereby helps advance the field toward more refined and context-aware security analysis aligned with the realistic characteristics of complex systems.