This paper presents a novel approach for detecting fraudulent tokens in blockchain transactions using an enhanced Graphs of Graphs model. By leveraging blockchain’s inherent transparency and immutability, our method enables precise tracking of transactional patterns, enhancing fraud detection capabilities and market forecast within decentralized ecosystems. We introduce a self-attention mechanism, Transaction-To-Token Attention (T2T-Attention), which captures both wallet-level and token-level information by maintaining detailed node and cluster interactions. Unlike traditional graph coarsening, T2T-Attention enables a fine-grained analysis of token transfer activities, preserving transaction-level details while also modeling inter-token relationships at a cluster level. To address computational complexity, we use kernelized softmax for efficient processing. Our model outperforms existing methods in detecting fraud and classifying token-related risks, offering a scalable solution to enhance blockchain security and DeFi risk management.

As the blockchain ecosystem continues to diversify, the lack of interoperability among heterogeneous blockchain systems has become a critical bottleneck, leading to fragmented data silos and limited collaboration. Although numerous cross-chain protocols—such as atomic swaps, sidechains, and relay-based mechanisms—have been introduced to address this issue, they often face significant challenges related to privacy, security, and decentralization. In this paper, we propose a novel cross-chain protocol that enhances traditional hash-locking mechanisms by integrating zero-knowledge proofs and chameleon hash functions. Our approach ensures strong path confidentiality, such that reconstructing the payment path is computationally infeasible under the discrete logarithm assumption, even in partially compromised networks. Additionally, we introduce a multi-path atomic swap framework that supports concurrent routing and preserves transactional autonomy, enabling users to flexibly select preferred payment paths. We evaluate the performance through theoretical analysis and simulation. Comparative results demonstrate that our solution achieves secure atomicity with minimal trust assumptions and improved latency compared to existing methods.

Blockchain-based identity brings in great evolution due to its decentralized deployment, transparent and tamper-free ledger. Specification groups of B5G/6G are exploring into integrate the technology to future network systems, e.g., Internet of Things, vehicular network, industrial communications. However, devices in these systems often have storage constraints and unstable channels, which necessitates lightweight node deployment. The security issue arises: revoked identity can forge a legitimate authentication, since the lightweight verifier does not maintain the revocation transactions. This paper hence proposes AcBF, a novel revocable identity management scheme, that enables extremely low authentication latency by allowing the lightweight node to query the certificate's status locally. To realize this feature trustfully, we design a revocation transaction based on accumulator-assisted Bloom filter to minimize the storage of certificate status structure. Secondly, we construct the blockchain protocol to ensure that no revocation event slips on any lightweight ledger, even in an insecure or unstable communication environment. In addition, different from other revocation mechanisms, AcBF minimizes the impact on valid users during the revocation process. Through security and performance analysis, AcBF has shown strong security and advantageous efficiency on both lightweight verifiers and certificate owners, thus suits identity management systems with low-latency constraints.

In the realm of graph learning, there is a category of methods that conceptualize graphs as hierarchical structures, utilizing node clustering to capture broader structural information. While generally effective, these methods often rely on a fixed graph coarsening routine, leading to overly homogeneous cluster representations and loss of node-level information. In this paper, we envision the graph as a network of interconnected node sets without compressing each cluster into a single embedding. To enable effective information transfer among these node sets, we propose the Node-to-Cluster Attention (N2C-Attn) mechanism. N2C-Attn incorporates techniques from Multiple Kernel Learning into the kernelized attention framework, effectively capturing information at both node and cluster levels. We then devise an efficient form for N2C-Attn using the cluster-wise message-passing framework, achieving linear time complexity. We further analyze how N2C-Attn combines bi-level feature maps of queries and keys, demonstrating its capability to merge dual-granularity information. The resulting architecture, Cluster-wise Graph Transformer (Cluster-GT), which uses node clusters as tokens and employs our proposed N2C-Attn module, shows superior performance on various graph-level tasks. Code is available at https://github.com/LUMIA-Group/Cluster-wise-Graph-Transformer.

Attention mechanisms have made significant strides in graph learning, yet they still exhibit notable limitations: local attention faces challenges in capturing long-range information due to the inherent problems of the message-passing scheme, while global attention cannot reflect the hierarchical neighborhood structure and fails to capture fine-grained local information. In this paper, we propose a novel multi-hop graph attention mechanism, named Subtree Attention (STA), to address the aforementioned issues. STA seamlessly bridges the fully-attentional structure and the rooted subtree, with theoretical proof that STA approximates the global attention under extreme settings. By allowing direct computation of attention weights among multi-hop neighbors, STA mitigates the inherent problems in existing graph attention mechanisms. Further we devise an efficient form for STA by employing kernelized softmax, which yields a linear time complexity. Our resulting GNN architecture, the STAGNN, presents a simple yet performant STA-based graph neural network leveraging a hop-aware attention strategy. Comprehensive evaluations on ten node classification datasets demonstrate that STA-based models outperform existing graph transformers and mainstream GNNs. The code is available at https://github.com/LUMIA-Group/SubTree-Attention.