检索结果-内蒙古大学图书馆

IEEE International Conference on Acoustics, Speech and Signal Processing

作者： Linhao Dong Shuang Xu Bo Xu Institute of Automation Chinese Academy of Sciences China

ISBN: (纸本)9781538646595

Recurrent sequence-to-sequence models using encoder-decoder architecture have made great progress in speech recognition task. However, they suffer from the drawback of slow training speed because the internal recurrence limits the training parallelization. In this paper, we present the Speech-Transformer, a no-recurrence sequence-to-sequence model entirely relies on attention mechanisms to learn the positional dependencies, which can be trained faster with more efficiency. We also propose a 2D-Attention mechanism, which can jointly attend to the time and frequency axes of the 2-dimensional speech inputs, thus providing more expressive representations for the Speech-Transformer. Evaluated on the Wall Street Journal (WSJ) speech recognition dataset, our best model achieves competitive word error rate (WER) of 10.9%, while the whole training process only takes 1.2 days on 1 GPU, significantly faster than the published results of recurrent sequence-to-sequence models.

关键词： Speech Recognition sequence-to-sequence Attention Transformer

来源：评论

学校读者我要写书评

暂无评论

Neural Chinese Word Segmentation as sequence to sequence Translation 6th

Neural Chinese Word Segmentation as Sequence to Sequence Tra...

引用

6th National Conference on Social Media Processing (SMP)

作者： Shi, Xuewen Huang, Heyan Jian, Ping Guo, Yuhang Wei, Xiaochi Tang, Yi-Kun Beijing Inst Technol Sch Comp Sci & Technol Beijing Engn Res Ctr High Volume Language Informa Beijing 100081 Peoples R China

ISBN: (纸本)9789811068058;9789811068041

Recently, Chinese word segmentation (CWS) methods using neural networks have made impressive progress. Most of them regard the CWS as a sequence labeling problem which construct models based on local features rather than considering global information of input sequence. In this paper, we cast the CWS as a sequence translation problem and propose a novel sequence-to-sequence CWS model with an attention-based encoder-decoder framework. The model captures the global information from the input and directly outputs the segmented sequence. It can also tackle other NLP tasks with CWS jointly in an end-to-end mode. Experiments on Weibo, PKU and MSRA benchmark datasets show that our approach has achieved competitive performances compared with state-of-the-art methods. Meanwhile, we successfully applied our proposed model to jointly learning CWS and Chinese spelling correction, which demonstrates its applicability of multi-task fusion.

关键词： Chinese word segmentation sequence-to-sequence Chinese spelling correction Natural language processing

来源：评论

学校读者我要写书评

暂无评论

Multi-Agent Trajectory Prediction With Spatio-Temporal sequence Fusion

引用

IEEE TRANSACTIONS ON MULTIMEDIA 2023年 25卷 13-23页

作者： Wang, Yu Chen, Shiwei Tongji Univ Coll Elect & Informat Engn Shanghai 201804 Peoples R China Bilibili Shanghai 200433 Peoples R China

Accurate trajectory prediction of surrounding agents is an important issue for building up an intelligent transportation system. Frequent interactions among agents have a major impact on their movement patterns. Current research mainly relies on agents' spatial structure associated with the last frame of the observation to model social interactions, while paying less attention to structure information from previous moments. In addition, existing methods merely consider temporal features of a single trajectory sequence, while neglecting temporal dependencies across multiple trajectories. In this work, we endeavor to capture comprehensively social interactions among agents with the proposed Spatio-Temporal sequence Fusion Network (STSF-Net). Specifically, we construct a spatio-temporal sequence that encodes contextual information taking explicitly spatial distributions of agents during movement into account while capturing socially temporal dependencies across multiple trajectory sequences. Besides, a social recurrent mechanism is introduced to explicitly capture temporal correlations between interactions by concerning spatial structure at each time-step. Finally, our model is evaluated on datasets covering pedestrian, vehicle, and heterogeneous multi-agent trajectories. Experimental evidence manifests that our method achieves excellent performance.

关键词： Multi-modal trajectory prediction saptio-temporal sequence fusion generative adversarial networks sequence-to-sequence

来源：评论

学校读者我要写书评

暂无评论

Constrained sequence-to-Tree Generation for Hierarchical Text Classification 22

Constrained Sequence-to-Tree Generation for Hierarchical Tex...

引用

45th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR)

作者： Yu, Chao Shen, Yi Mao, Yue Alibaba Grp Beijing Peoples R China

ISBN: (纸本)9781450387323

Hierarchical Text Classification (HTC) is a challenging task where a document can be assigned to multiple hierarchically structured categories within a taxonomy. The majority of prior studies consider HTC as a flat multi-label classification problem, which inevitably leads to "label inconsistency" problem. In this paper, we formulate HTC as a sequence generation task and introduce a sequence-to-tree framework (Seq2Tree) for modeling the hierarchical label structure. Moreover, we design a constrained decoding strategy with dynamic vocabulary to secure the label consistency of the results. Compared with previous works, the proposed approach achieves significant and consistent improvements on three benchmark datasets.

关键词： text representation text classification sequence-to-sequence

来源：评论

学校读者我要写书评

暂无评论

Sleep Staging Using Compressed Vision Transformer With Novel Two-Step Attention Weighted Sum

引用

IEEE ACCESS 2025年 13卷 69650-69659页

作者： Kim, Hyounggyu Kim, Moogyeong Chung, Wonzoo Korea Univ Dept Artificial Intelligence Seoul 02841 South Korea

Automatic sleep staging is crucial for diagnosing sleep disorders, however, existing inter-epoch feature extraction schemes such as RNN-based networks or transformers often struggle with long sleep sequences due to overfitting. This study presents a novel automatic sleep staging method utilizing a pre-trained vision transformer with compression as a sequence encoder and a two-step attention to enhance the sleep-stage classification performance. In contrast to existing transformer-based methods, the pre-trained transformer with compression can handle long sequences covering a sleep cycle, leveraging robust feature extraction capabilities with substantially fewer parameters. Furthermore, an epoch encoder based on a bidirectional temporal convolutional network with a multi-head two-step attention mechanism is proposed to improve the efficiency of epoch-level feature extraction. The performance of the proposed method is evaluated using three publicly available datasets: SleepEDF-20, SleepEDF-78, and SHHS. Numerical experiments show notable performance enhancement of the proposed scheme in comparison with the state-of-the-art algorithms, particularly for small training datasets, which validates the resilience of the proposed method against overfitting. These results suggest that with appropriate regularization, transformer-based models can effectively capture long-term contextual information across a complete sleep cycle.

关键词： Sleep Transformers Electroencephalography Feature extraction Recording Time-frequency analysis Computer vision Brain modeling Overfitting Bidirectional long short term memory Automatic sleep staging electroencephalogram long-term dependency modeling sequence-to-sequence vision transformer temporal convolutional network

来源：评论

学校读者我要写书评

暂无评论

Transfer learning-based energy consumption prediction for variable refrigerant flow system in buildings

引用

APPLIED THERMAL ENGINEERING 2025年 267卷

作者： Park, Chanwoo Kim, Icksung Kim, Woohyun Chonnam Natl Univ Sch Mech Engn Gwangju South Korea

Accurately predicting the energy consumption of variable refrigerant flow (VRF) systems is vital for effective energy management. However, deep learning models trained on one VRF system cannot generalize easily to other VRF systems, which limits the scalability of deep learning for VRF systems. Transfer learning offers a promising solution to this problem by enabling predictions for systems with limited data due to the absence of monitoring systems or a lack of data collection time. In this study, we developed a transfer learning model based on a sequence-to-sequence long short-term memory architecture and trained it on a two-week energy consumption dataset. We explored various strategies to determine the number of layers that should be frozen, finetuned, and added based on the target data for VRF systems. Different Bayesian optimization techniques and similarity measures were considered to enhance the accuracy of the model and reduce its training time. Compared with that of the traditional deep learning model, the coefficient of variation of the root mean squared error for the transfer learning model was 6-19% lower. Additionally, hyperparameter optimization during the pre-training process reduced the computation time by up to 55%. When integrated into a building energy management system, the optimized transfer learning model could accurately predict the energy consumption of buildings equipped with VRF systems in real-time.

关键词： Variable refrigerant flow Transfer learning sequence-to-sequence Similarity measure Bayesian optimization

来源：评论

学校读者我要写书评

暂无评论

Predicting Morphological Changes Along a Macrotidal Coastline Using a Two-Stage Machine Learning Model

引用

WATER RESOURCES RESEARCH 2025年第4期61卷

作者： Kumar, Pavitra Leonardi, Nicoletta Univ Liverpool Sch Environm Sci Dept Geog & Planning Liverpool England

Understanding and predicting coastal change is of the foremost importance to protect coastal communities and coastal assets. This study analyzes field data from 125 locations along the Morecambe coastline, consisting of beach transects collected twice a year for more than a decade (2007-2022). Wave data at these 125 locations were simulated using the hydrodynamic Delft3D model, with full coupling of the Delft3D FLOW and WAVE modules. To model the sediment volume changes observed along the Morecambe coastline, this study proposes a two-stage machine learning model that incorporates beach behavior classification and deep learning techniques to predict changes in sediment volumes along coastal environments. The first stage of the model, developed using a random forest classifier, classifies beach behavior into four categories: eroding, accreting, stable, or undergoing short-term fluctuations. The second stage of the model developed using LSTM and sequence-to-sequence models, uses the output of the first stage to predict the change in sediment volume after erosion/accretion. The random forest classifier achieves testing accuracy of 0.74. LSTM model achieved a testing regression of 0.92 for one-step-ahead (6 months) predictions of change in sediment volume time series, while sequence-to-sequence model achieved the testing regression of 0.96 for three-time-ahead (1.5 years) predictions and 0.88 for ten-time-step-ahead (5 years) prediction.

关键词： sediment volume morecambe bay two-stage modeling random forest LSTM sequence-to-sequence

来源：评论

学校读者我要写书评

暂无评论

Decomposition of high-density sEMG signals: Extracting multiple spikes from single time windows

引用

BIOMEDICAL SIGNAL PROCESSING AND CONTROL 2025年 107卷

作者： Jiang, Xi Guo, Weiyu Cui, Ziwei Lin, Chuang Su, Jingyong Harbin Inst Technol Shenzhen Sch Comp Sci & Technol Shenzhen Peoples R China Hong Kong Univ Sci & Technol Thrust Artificial Intelligence Guangzhou Peoples R China Dalian Maritime Univ Sch Informat Sci & Technol Dalian Peoples R China

Decomposing high-density surface electromyography (HD-sEMG) signals has become a powerful tool in various applications, including prosthetic control and human-machine interaction (HMI). Using deep learning methods to decompose HD-sEMG signals can eliminate preprocessing such as whitening of sEMG, thereby reducing its latency in HMI applications. However, current deep learning methods for blind separation mainly use window-based classification methods, which cannot accurately decompose spike-dense areas. In this paper, we rethink from the perspective of sequence to sequence (seq2seq), define the surface electromyography signal decomposition problem as a regression problem, and propose an HD-sEMG signal decoding method CKC-TCN. This is the first time that the problem of extracting multiple spikes from single time windows has been solved. Rather than class labels, we treat the innervation pulse trains (IPTs) of each motor unit (MU) that are derived from the convolution kernel compensation (CKC) algorithm as continuous time series. We train the temporal convolutional network (TCN) to extract sample-level accuracy IPTs of each MU from the unprocessed HD-sEMG signals, and then extract MU firing time sequences. To evaluate the effectiveness of the proposed method, we conduct experiments on both simulated and real data of HD-sEMG. The results show that, CKC-TCN reduces inference time by 99% compared to traditional methods and improves separation accuracy by over 10% compared to the state-of-the-art deep learning solutions, achieving a significant performance enhancement. This makes it more suitable for real-time applications.

关键词： HD-sEMG Motor unit Blind source separation sequence-to-sequence Temporal convolutional network

来源：评论

学校读者我要写书评

暂无评论

Investigate How Market Behaves: Toward an Explanatory Multitasking Based Analytical Model for Financial Investments

引用

IEEE ACCESS 2024年 12卷 30928-30940页

作者： Das, Sarmistha Chowdhury, Ujjwal Lijin, N. S. Deep, Atulya Saha, Sriparna Maurya, Alka IIT Patna CSE Dept Patna 801106 India Ramakrishna Mission Vivekananda Educ & Res Inst Belur 711202 West Bengal India IISER Bhopal Bhopal 462066 Madhya Pradesh India SRM Inst Sci & Technol Chennai 603203 India Crisil Pvt Ltd Mumbai 400076 India

In recent years, the intersection between financial market behavior and social media has emerged as a sought-after source of information, meeting the requirements of investors, institutions, regulators, researchers, and policymakers. Assessing sentiment and emotions aids in evaluating public psychology on particular stocks, assets, or the overall market, with shifts often aligning with market movements. Previously, machine learning, both traditional and deep learning methods, targeted discerning stock market sentiment and emotion without conducting studies to offer comprehensive explanations for these behavioral factors. In this study, we introduce a multitasking sequence-to-sequence model that integrates financial investment analysis with sentiment and emotion analysis from tweets upheld by an explanation mechanism. We also present the FinEMA dataset, featuring sentiment, emotion, and cause labels on financial stock market changes. Our study highlights how joint learning improves performance in discerning sentiment and emotion by utilizing interrelated features, enhancing task effectiveness. Our proposed model, the Emotion-Sentiment Attention Network (ESAN), achieved 89% accuracy in sentiment identification and 79% accuracy in emotion recognition, outperforming conventional machine learning methods. Furthermore, our findings indicate a positive outlook for the stock market in the latter half of 2023, which has intensified investor optimism, though some individuals still harbor uncertainties. Conclusively, our results suggest that regenerating existing computational tools can open up new research opportunities to address relevant novel tasks. The primary aim of this study is to elucidate the diverse dimensions of financial market behaviour and offer explanatory insights for the research community. The authors maintain impartiality towards specific stocks. It's essential to note that stock market investments inherently carry market risks and potential losses. The mar

关键词： Investment Social networking (online) Stock markets Task analysis Emotion recognition Sentiment analysis Multitasking Financial services Sequential analysis Market research Financial market emotion and sentiment analysis explainability multi-tasking sequence-to-sequence social sentiment on the stock market

来源：评论

学校读者我要写书评

暂无评论

SeqESR-GAN-Based Sparse Data Augmentation for Distribution Networks

引用

IEEE TRANSACTIONS ON INDUSTRIAL INFORMATICS 2024年第11期20卷 12913-12923页

作者： Xu, Tao Zhang, Jiadong Meng, He Liu, Lutong Wang, Kaiqi Qiao, Ji Zhao, Zixuan Zhu, Hong Wang, Wendi Tianjin Univ Key Lab Smart Grid Minist Educ Tianjin 300072 Peoples R China China Elect Power Res Inst Beijing 100192 Peoples R China State Grid Nanjing Elect Power Co Nanjing 210024 Peoples R China

The increasing integration of distributed energy resources brings considerable uncertainties to power distribution systems. Limited redundancy in measurement not only constrains dynamic state estimation but also inevitably overlooks the electricity generation/utilization characteristics of typical prosumers. To enhance the system observability, a three-stage sparse data augmentation framework, namely sequence-to-sequence enhanced super-resolution generative adversarial network (GAN) is established. A novel data image encoding method is proposed to reflect the periodic electricity utilization patterns of renewable energy sources, loads, and energy storage systems, and enhances measurement data resolution by learning energy flow behaviors. Different learning strategies are employed in the initial two stages, enabling the GAN-based model to capture the spatiotemporal characteristics of the energy tensor and restore high-frequency elements. The third stage involves a gated recurrent unit-based Seq2Seq model to eliminate time lags and invalid details in the super-resolved data. Case studies and analyses are carried out to validate the effectiveness of the proposed approach.

关键词： Superresolution Image reconstruction Tensors Generative adversarial networks Power systems Electricity Data augmentation Distribution network generative adversarial network (GAN) sequence-to-sequence super-resolution reconstruction

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：