Pedestrian trajectory prediction is a key component for various applications that involve human and vehicle interactions, such as autonomous driving, traffic management and smart city planning. Existing methods based ...
详细信息
Pedestrian trajectory prediction is a key component for various applications that involve human and vehicle interactions, such as autonomous driving, traffic management and smart city planning. Existing methods based on graph neural networks have limited ability to capture group interactions and precisely model complex associations among multi-agents. To solve these problems, we propose OST-HGCN, an optimized hypergraph convolutional network. It models multi-agent trajectory interactions from both temporal and spatial perspectives using hypergraphstructures, and optimizes the spatio-temporal hypergraphstructure to enable fine-grained analysis of multi-agent trajectory motion intentions and high-order interactions. We employ OST-HGCN to a CVAE-based prediction framework, and use the optimized hypergraphstructure to predict multi-agent plausible trajectories. We conduct extensive experiments on four real trajectory prediction datasets of NBA, NFL, SDD and ETH-UCY, and verify the effectiveness of the proposed OST-HGCN.
Multimodal Sentiment Analysis (MSA) is the process of relying on multimodal information, such as text, audio, and visual, to determine a subject's affective tendencies. While many recent studies have adopted graph...
详细信息
Multimodal Sentiment Analysis (MSA) is the process of relying on multimodal information, such as text, audio, and visual, to determine a subject's affective tendencies. While many recent studies have adopted graph-based techniques for MSA, they have yet to fully explore the sentimental interactions both within unimodal temporal steps and across individual modalities. To address the limitations arising from the isolation of unimodal hypergraphs in affective relationship mining, this paper proposes a multimodal hypergraph network based on contrastive learning. It constructs the hypergraphstructure by utilizing the sequential time steps of all three modalities as a collection of nodes, aiming to explore the multidimensional affective relationships across uni-, bi-, and tri-modalities. Specifically, this paper first generates the initial hypergraphstructure using a correlation-based hypergraph construction method to ensure the effectiveness of the constructed hypergraph. Then, both supervised and unsupervised contrastive learning methods are designed to optimize feature learning and the structure of the multimodal hypergraph, adaptively and simultaneously capturing the relationships among the time-series nodes of uni-, bi-, and tri-modalities. The proposed methods in this paper have demonstrated the advantages and effectiveness by conducting a large number of comparative and validation experiments on the English CMU-MOSI and CMU-MOSEI datasets as well as the Chinese CH-SIMS dataset.
Generative AI is revolutionizing Software Engineering (SE), as both engineers and academics embrace this technology in their work. To better leverage this technology for software generation, it is essential to propose...
详细信息
Generative AI is revolutionizing Software Engineering (SE), as both engineers and academics embrace this technology in their work. To better leverage this technology for software generation, it is essential to propose effective IoT software defect prediction methods. However, this task is challenging due to the unclear high-order correlation underlying the data. Moreover, in real-world IoT software defect prediction applications, different types of misclassifications generally lead to distinct losses and associated costs. However, accurately determining these specific costs is often not feasible. Under such circumstances, we propose a cost-sensitive hypergraph learning method with structure quality preservation (csHLQ) to optimize the cost information and preserve the graph quality in a principled way. Due to the representational ability on high-order relationship exploring, we employ hypergraphstructure instead of graph structure to model the complex correlations among the datasets. We note that if a cost-sensitive hypergraph has a high quality, its classification results may exhibit a large margin separation. Thus, csHLQ exploits the large margin cost-sensitive hypergraph while avoiding using of a cost-sensitive hypergraph with a small margin. To measure the performance of our proposed method, we performed experiments on three distinct groups of datasets, i.e., the NASA Metrics Data Program (NASA) dataset, CK metric dataset and UCI Machine Learning Repository (UCI) dataset. Experimental results and comparisons with state-of-the-art methods demonstrate the superiority of our method.
暂无评论