Detecting Argument-Swap Bugs Using Context-Enhanced Code Representations
本文提出了一种名为BugProbe的学习方法，通过结合上下文信息和基于名称的相似性来检测Python方法调用中参数顺序错误的问题，无需依赖于调用到定义的映射。
0 citationsRead paper
Selected work
Representative Papers
Enhancing Reliability of Symbolic Execution Tools for Smart Contract Analysis through Rule-Based False Positive Reduction
This study addresses the high false positive rate of Mythril, a symbolic execution tool for Ethereum smart contract vulnerability detection. It presents the first systematic analysis of the root causes behind these false positives and introduces a novel rule-based filtering mechanism. By integrating symbolic execution with static analysis and employing tailored rules, the proposed approach accurately distinguishes genuine vulnerabilities from spurious reports across major vulnerability categories. Experimental results demonstrate that the method substantially reduces false positives without compromising detection completeness, thereby enhancing both the reliability and practical utility of Mythril in real-world smart contract analysis.
0 citationsRead paper
Quantum Enchanced Multi-Scale CNN with Bi-directional Mamba for Crop Field Analysis
This study addresses the challenges of hyperspectral image-based crop classification, which include high-dimensional spectral data, complex spatial structures, class imbalance, and limited labeled samples. To tackle these issues, the authors propose the BiSpectral Mamba framework, which uniquely integrates a bidirectional Mamba state space model with quantum-inspired learning, while also incorporating multiscale CNNs and a spectral attention mechanism to enable efficient joint spatial–spectral feature extraction. A class-weighted training strategy is further employed to enhance classification robustness under few-shot conditions. Evaluated on the UAVHSI-Crop dataset, the method achieves an overall accuracy of 84.83%, demonstrating its strong potential for agricultural remote sensing applications.
0 citationsRead paper
xModel-KD: Cross-modal Knowledge Distillation for 3D Scene Perception using LiDAR
This work addresses the high annotation cost and limitations of single-modality perception in 3D point cloud segmentation by proposing a cross-modal knowledge distillation framework that effectively integrates 2D image texture cues with 3D geometric structures to learn unified per-point representations. Leveraging pre-trained 2D and 3D backbone networks, a cross-modal fusion encoder, and a multi-view contrastive alignment objective, the method enables efficient dense prediction without requiring extensive labeled data. Experimental results demonstrate that the approach achieves an absolute improvement of 2% in mean Intersection over Union (mIoU) over LiDAR-only baselines, validating its effectiveness and scalability for 3D scene understanding.
0 citationsRead paper
ConTrans: Learning Text-enhanced Local-global Temporal Representations for Zero-shot Temporal Action Localization
This work addresses the limitations in zero-shot temporal action localization, particularly the neglect of local relative frame offsets and insufficient feature representation capacity. To overcome these issues, the authors propose ConTrans, a novel architecture that, for the first time, jointly models local relative offsets and global semantics. ConTrans integrates convolutional inductive bias with Transformer self-attention, employing a multi-scale encoder to simultaneously capture fine-grained local dependencies and long-range global context. Furthermore, it introduces text-enhanced learning of local-global temporal representations. Extensive experiments on ActivityNet-1.3 and THUMOS14 demonstrate that ConTrans significantly outperforms existing methods, establishing new state-of-the-art performance in zero-shot temporal action localization.
0 citationsRead paper
Recent publications
Latest Papers
Detecting Argument-Swap Bugs Using Context-Enhanced Code Representations
本文提出了一种名为BugProbe的学习方法，通过结合上下文信息和基于名称的相似性来检测Python方法调用中参数顺序错误的问题，无需依赖于调用到定义的映射。
0 citationsRead paper
Enhancing Reliability of Symbolic Execution Tools for Smart Contract Analysis through Rule-Based False Positive Reduction
This study addresses the high false positive rate of Mythril, a symbolic execution tool for Ethereum smart contract vulnerability detection. It presents the first systematic analysis of the root causes behind these false positives and introduces a novel rule-based filtering mechanism. By integrating symbolic execution with static analysis and employing tailored rules, the proposed approach accurately distinguishes genuine vulnerabilities from spurious reports across major vulnerability categories. Experimental results demonstrate that the method substantially reduces false positives without compromising detection completeness, thereby enhancing both the reliability and practical utility of Mythril in real-world smart contract analysis.
0 citationsRead paper
Quantum Enchanced Multi-Scale CNN with Bi-directional Mamba for Crop Field Analysis
This study addresses the challenges of hyperspectral image-based crop classification, which include high-dimensional spectral data, complex spatial structures, class imbalance, and limited labeled samples. To tackle these issues, the authors propose the BiSpectral Mamba framework, which uniquely integrates a bidirectional Mamba state space model with quantum-inspired learning, while also incorporating multiscale CNNs and a spectral attention mechanism to enable efficient joint spatial–spectral feature extraction. A class-weighted training strategy is further employed to enhance classification robustness under few-shot conditions. Evaluated on the UAVHSI-Crop dataset, the method achieves an overall accuracy of 84.83%, demonstrating its strong potential for agricultural remote sensing applications.
0 citationsRead paper
xModel-KD: Cross-modal Knowledge Distillation for 3D Scene Perception using LiDAR
This work addresses the high annotation cost and limitations of single-modality perception in 3D point cloud segmentation by proposing a cross-modal knowledge distillation framework that effectively integrates 2D image texture cues with 3D geometric structures to learn unified per-point representations. Leveraging pre-trained 2D and 3D backbone networks, a cross-modal fusion encoder, and a multi-view contrastive alignment objective, the method enables efficient dense prediction without requiring extensive labeled data. Experimental results demonstrate that the approach achieves an absolute improvement of 2% in mean Intersection over Union (mIoU) over LiDAR-only baselines, validating its effectiveness and scalability for 3D scene understanding.
0 citationsRead paper
ConTrans: Learning Text-enhanced Local-global Temporal Representations for Zero-shot Temporal Action Localization
This work addresses the limitations in zero-shot temporal action localization, particularly the neglect of local relative frame offsets and insufficient feature representation capacity. To overcome these issues, the authors propose ConTrans, a novel architecture that, for the first time, jointly models local relative offsets and global semantics. ConTrans integrates convolutional inductive bias with Transformer self-attention, employing a multi-scale encoder to simultaneously capture fine-grained local dependencies and long-range global context. Furthermore, it introduces text-enhanced learning of local-global temporal representations. Extensive experiments on ActivityNet-1.3 and THUMOS14 demonstrate that ConTrans significantly outperforms existing methods, establishing new state-of-the-art performance in zero-shot temporal action localization.
0 citationsRead paper