The document-level relation extraction task aims to extract relational triples from a document consisting of multiple sentences. Most previous models focus on modeling the dependency between the entities and neglect t...
详细信息
ISBN:
(纸本)9781450395656
The document-level relation extraction task aims to extract relational triples from a document consisting of multiple sentences. Most previous models focus on modeling the dependency between the entities and neglect the reasoning mechanism. Some other models construct paths implicitly between co-sentence entities to find semantic relations. However, they ignore that there are interactions between different triples, especially some triples play an import role in predicting others. In this short research paper, we propose a new two stage framework PCSR(Pre-classification Supporting Reasoning) which captures the interactions between triples and utilizes these information for reasoning. Specifically, we make a pre-classification for each entity pair in the first stage. Then we aggregate the embeddings of predicted triples to enhance entity representation and make a new classification. Since the second classification could find triples missed in the first stage, we take the result as the supplement of the prior one. Experiments on DocRED show that our method achieves an F1 score of 62.11. Compared with the previous state-of-the-art model, our model increase by 0.81 on the test set, which demonstrates the effectiveness of our reasoning mechanism.
This paper explores the size-invariance of evaluation metrics in Salient Object Detection (SOD), especially when multiple targets of diverse sizes co-exist in the same image. We observe that current metrics are size-s...
详细信息
Self-supervised learning (SSL) recently has achieved outstanding success on recommendation. By setting up an auxiliary task (either predictive or contrastive), SSL can discover supervisory signals from the raw data wi...
详细信息
This paper develops an end-to-end ECG signal classification algorithm based on a novel segmentation strategy and 1D Convolutional Neural Networks (CNN) to aid the classification of ECG signals and alleviate the worklo...
详细信息
This paper develops an end-to-end ECG signal classification algorithm based on a novel segmentation strategy and 1D Convolutional Neural Networks (CNN) to aid the classification of ECG signals and alleviate the workload of physicians. The ECG segmentation strategy named R-R-R strategy (i.e., retaining ECG data between the R peaks just before and after the current R peak) is used for segmenting the original ECG data into segments to train and test the 1D CNN models. The novel strategy mimics physicians in scanning ECG to a greater extent, and maximizes the inherent information of ECG segments for diagnosis. The performance of the proposed end to end ECG signal classification algorithm was verified with the ECG signals from 48 records in the MIT-BIH arrhythmia database. When the heartbeat types were divided into the five classes recommended by clinicians, i.e., normal beat, left bundle branch block beat, right bundle branch block beat, premature ventricular contraction, and paced beat, the classification accuracy, the area under the curve (AUC), the sensitivity, and the F1-score achieved by the proposed model were 0.9924, 0.9994, 0.99 and 0.99, respectively. When the heartbeat types were divided into six classes recommended by clinicians, i.e., normal beat, left bundle branch block beat, right bundle branch block beat, premature ventricular contraction, paced beat and other beats, the beat classification accuracy, the AUC, the sensitivity, and the F1-score achieved by the model reached 0.9702, 0.9966, 0.97, and 0.97, respectively. When the heartbeat types were divided into five classes recommended by the Association for Advancement of Medical Instrumentation (AAMI), i.e., normal beat, supraventricular ectopic beat, ventricular ectopic beat, fusion beat, and unknown beat, the beat classification accuracy, the sensitivity, and the F1-score were 0.9745, 0.97, and 0.97, respectively. Experimental results show that the proposed method achieves better performance than the s
In real-world scenarios, scanned point clouds are often incomplete due to occlusion issues. The tasks of self-supervised and weakly-supervised point cloud completion involve reconstructing missing regions of these inc...
详细信息
Depression can significantly impact many aspects of an individual’s life, including their personal and social functioning, academic and work performance, and overall quality of life. Many researchers within the field...
详细信息
Clustering analysis is important for providing personalized services across various user experience optimization. As an advanced clustering technique, Fuzzy Co-Clustering (FCC) excels at handling high-dimensional data...
详细信息
Clustering analysis is important for providing personalized services across various user experience optimization. As an advanced clustering technique, Fuzzy Co-Clustering (FCC) excels at handling high-dimensional data by simultaneously grouping both samples and attributes while allowing objects to belong to multiple clusters. To alleviate the potential privacy risks posed by directly using raw data, researchers often employ local differential privacy (LDP) techniques for protection. However, current clustering studies under LDP either directly perform clustering on the high-dimensional data, which necessitates splitting the privacy budget heavily, leading to excessive noise injection, or fail to adequately address the impact of Laplacian noise on both objective function modeling and distance metric selection, resulting in reduced clustering accuracy. To address these limitations, we present a two-phase approach SONY, which allocates part of the privacy budget to evaluate attribute importance and perform dimensionality reduction accordingly, and then utilizes the remaining budget to conduct noise-aware clustering on the pruned data. Our key technical contributions include: (1) a Noise-aware Similarity Calculation method (NSC) that replaces traditional Euclidean distance with a specialized metric which accounts for the statistical properties of Laplacian noise;(2) an importance-driven dimension reduction and budget allocation method (IDRBA) which optimizes both attribute selection and privacy budget allocation;and (3) an outlier-aware objective function calculation method (OOC) which enhances robustness against potential noise-induced outliers. We provide theoretical guarantees for SONY's convergence, privacy preservation, utility bounds, and computational complexity. Experimental evaluations conducted on two real-world datasets and one synthetic dataset, demonstrate that SONY outperforms the state-of-the-arts by at least 14% in F-Measure and 10% in Entropy while main
Reconfigurable intelligent surface (RIS) and ambient backscatter communication (AmBC) technologies are recognized for their programmability and high energy efficiency respectively, which will be the key parts of the f...
详细信息
Multi-view clustering based on non-negative matrix factorization (NMFMvC) is a well-known method for handling high-dimensional multi-view data. To satisfy the non-negativity constraint of the matrix, NMFMvC is usually...
详细信息
The growing demand for location-based services in areas like virtual reality, robot control, and navigation has intensified the focus on indoor localization. Visible light positioning (VLP), leveraging visible light c...
详细信息
暂无评论