Article
Computer Science, Artificial Intelligence
Yongping Du, Yang Liu, Zhi Peng, Xingnan Jin
Summary: Sentiment classification is important for helping people make better decisions by exploring their expressed opinions. This paper introduces a novel multimodal sentiment classification model based on a gated attention mechanism. The model emphasizes text segments by using image features and focuses on the text that affects sentiment polarity. Experimental results demonstrate the effectiveness of the proposed model in outperforming previous state-of-the-art models.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Chao Yang, Xianzhi Wang, Lina Yao, Guodong Long, Jing Jiang, Guandong Xu
Summary: Multivariate time series classification is a critical problem in data mining with broad applications. We propose a novel convolutional neural network architecture called Attentional Gated Res2Net for multivariate time series classification. Our model utilizes hierarchical residual-like connections to capture multi-granular temporal information and two types of attention modules for better leveraging the temporal patterns. Experimental results demonstrate that our model outperforms several baselines and state-of-the-art methods on 14 benchmark multivariate time-series datasets.
NEURAL PROCESSING LETTERS
(2023)
Article
Computer Science, Hardware & Architecture
Qingchun Bai, Jie Zhou, Liang He
Summary: A Position-Gated Recurrent Neural Networks (PG-RNN) model is proposed for aspect-based sentiment classification, which dynamically integrates global and local information and achieves significant improvement in accuracy. The model considers aspect word position information and utilizes a positional RNN model and representation absorption gating for enhanced performance.
JOURNAL OF SUPERCOMPUTING
(2022)
Article
Computer Science, Artificial Intelligence
Mustafa Kaytan, Ibrahim Berkan Aydilek, Celaleddin Yeroglu
Summary: In CNNs, the choice of activation functions is crucial. Recent studies have shown that non-monotonic activation functions like Swish, Mish, Logish, and Smish outperform traditional ones like ReLU. In this study, a new activation function called Gish is proposed, which exhibits superior performance compared to existing functions. Experimental results on different network models and datasets confirm the effectiveness of Gish.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Daming Li, Lianbing Deng, Zhiming Cai
Summary: A new convolution kernel proposed in this study can detect corresponding features with different transformations by actively changing the positions of connections, improving the image classification effect. Replacing a traditional convolution kernel with a complex one significantly enhances network performance.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Mustafa Kaytan, Ibrahim Berkan Aydilek, Celaleddin Yeroglu
Summary: The importance of selecting and using appropriate activation functions in CNNs is discussed in this paper. A new activation function called Gish is proposed, and its effectiveness is demonstrated through experiments on different network models and datasets.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Computer Science, Interdisciplinary Applications
Xuefen Lin, Jielin Chen, Weifeng Ma, Wei Tang, Yuchen Wang
Summary: This study proposes an improved graph convolution model that achieves effective emotion classification in complex dataset environments and reduces the cost of affective computing.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2023)
Article
Computer Science, Artificial Intelligence
Zhaoyang Deng, Chenxiang Sun, Guoqiang Zhong, Yuxu Mao
Summary: In this paper, a novel graph-based model is proposed, where each document is represented as a text graph. The semantic information of each word node is propagated and updated with an attention gated graph neural network (AGGNN) and keyword nodes with discriminative semantic information are extracted via an attention-based text pooling layer (TextPool), transforming text classification into a graph classification task.
COGNITIVE COMPUTATION
(2022)
Article
Chemistry, Multidisciplinary
Zhiwei Liu, Ting Bian, Minglai Yang
Summary: This paper proposes a novel approach, LGNet, to address the issue of large intra-class differences in music genre classification. By incorporating multiple locally activated multi-layer perceptrons and a gated routing network, LGNet adaptively learns from music signals with diverse characteristics. Experimental results demonstrate that LGNet outperforms existing methods, achieving superior performance on the filtered GTZAN dataset.
APPLIED SCIENCES-BASEL
(2023)
Article
Environmental Sciences
Deen Dai, Lihua Cao, Yangfan Liu, Yao Wang, Zhaolong Wu
Summary: This study proposes a high-altitude flying object classification algorithm based on radiation characteristic data. The algorithm utilizes infrared detection technology to obtain target images and performs classification using attention-based convolutional neural networks and gated recurrent units. It achieves high classification accuracy and F1 score.
Article
Computer Science, Information Systems
Nayan Kumar Sarkar, Moirangthem Marjit Singh, Utpal Nandi
Summary: The paper introduces a Deep Learning-based approach for crop disease classification, using network deconvolution operation and attention-based activation function. The results show that the proposed model, utilizing network deconvolution operation and AReLU activation function, significantly outperforms other existing models in crop disease classification.
MULTIMEDIA TOOLS AND APPLICATIONS
(2023)
Article
Mathematical & Computational Biology
Huang Chengcheng, Yuan Jian, Qin Xiao
Summary: With the rapid development of apparel e-commerce, the classification of apparel based on its collar design has become increasingly important. Traditional image processing methods struggle with complex image backgrounds. To address the issue, an EMRes-50 classification algorithm is proposed, which combines the ECA-ResNet50 model with the MC-Loss loss function method. The algorithm achieved high classification accuracy when applied to the Coller-6 and DeepFashion-6 datasets. Experimental results demonstrate that the improved model outperforms existing CNN models in terms of accuracy and feature extraction ability.
FRONTIERS IN COMPUTATIONAL NEUROSCIENCE
(2022)
Article
Computer Science, Artificial Intelligence
J. Ashok Kumar, S. Abirami, Tina Esther Trueman, Erik Cambria
Summary: Toxicity identification is a serious issue in online communities, and an automatic system like MCBiGRU is proposed for detecting toxic comments. Experimental results show that the MCBiGRU model outperforms in terms of multilabel metrics.
Article
Geochemistry & Geophysics
Xiao Li, Lin Lei, Yuli Sun, Ming Li, Guangyao Kuang
Summary: This study compares the complementary effects between optical and SAR features in land cover classification, proposing a novel collaborative attention-based fusion network that hierarchically fuses both types of features. The network utilizes multi-stage feature learning, collaborative attention mechanisms, and a gated fusion module to automatically learn the varying contributions of optical and SAR features, showing advantages over existing methods in land cover classification.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2021)
Article
Engineering, Electrical & Electronic
Shaohui Ning, Yansong Wang, Wenan Cai, Zhenlin Zhang, Yukun Wu, Yonglei Ren, Kangning Du
Summary: This study addresses the fault diagnosis problem of rolling bearings and improves the ShufflenetV2 network by sequentially embedding LSTM and Dropout layers to enhance accuracy and generalization ability. Experimental analysis shows that the proposed algorithm achieves high accuracy in real-time identification of early rolling bearing failures.
JOURNAL OF SENSORS
(2022)