检索结果-内蒙古大学图书馆

2024 Spoken Language Technology Workshop

作者： Wagner, Dominik Baumann, Ilja Ranzenberger, Thomas Riedhammer, Korbinian Bocklet, Tobias TH Nurnberg Georg Simon Ohm Nurnberg Germany

ISBN: (纸本)9798350392265;9798350392258

We present a method to personalize large transformer-based encoder-decoder speech foundation models without the need for changes in the underlying model structure or training from scratch. This is achieved by projecting speaker-specific information into the latent space of the transformer decoder via a small neural network and learning to process the speaker information along with domain-specific information via parameter-efficient finetuning. We use this method to improve the automatic speech recognition results of spoken academic German and English. Our approach yields average relative word error rate (WER) improvements of approximately 29% on German academic speech and 25% on English academic speech. It also translates well to conversational speech, achieving relative WER improvements of up to 36%, and demonstrates modest gains of up to 5% on read speech. Moreover, we observe that incorporating utterances from the recent past as personalization context yields the most significant overall improvements and that changes in voice characteristics resulting from prolonged speaking have a minimal effect on the personalization quality of academic lectures.

关键词： automatic speech recognition speaker adaptation personalization adalora encoder-decoder

来源：评论

学校读者我要写书评

暂无评论

MGB-Unet: An Improved Multiscale Unet with Bottleneck Transformer for Myositis Segmentation from Ultrasound Images

引用

JOURNAL OF IMAGING INFORMATICS IN MEDICINE 2025年第1期38卷 217-228页

作者： Hussein, Allaa Youssef, Sherin Ahmed, Magdy A. Ghatwary, Noha Pharos Univ Comp Engn Pharos Egypt Arab Acad Sci & Technol Comp Engn Alexandria Egypt Fac Engn Comp Engn Alexandria Egypt

Myositis is the inflammation of the muscles that can arise from various sources with diverse symptoms and require different treatments. For treatment to achieve optimal results, it is essential to obtain an accurate diagnosis promptly. This paper presents a new supervised segmentation architecture that can efficiently perform precise segmentation and classification of myositis from ultrasound images with few computational resources. The architecture of our model includes a unique encoder-decoder structure that integrates the Bottleneck Transformer (BOT) with a newly developed Residual block named Multi-Conv Ghost switchable bottleneck Residual Block (MCG_RB). This block effectively captures and analyzes ultrasound image input inside the encoder segment at several resolutions. Moreover, the BOT module is a transformer-style attention module designed to bridge the feature gap between the encoding and decoding stages. Furthermore, multi-level features are retrieved using the MCG-RB module, which combines multi-convolution with ghost switchable residual connections of convolutions for both the encoding and decoding stages. The suggested method attains state-of-the-art performance on a benchmark set of myositis ultrasound images across all parameters, including accuracy, precision, recall, dice coefficient, and Jaccard index. Despite its limited training data, the suggested approach demonstrates remarkable generalizability by yielding exceptional results. The proposed model showed a substantial enhancement in accuracy when compared to segmentation state-of-the-art methods such as Unet++, DeepLabV3, and the Duck-Net. The dice coefficient and Jaccard index obtained improvements of up to 3%, 6%, and 7%, respectively, surpassing the other methods.

关键词： Myositis Segmentation Unet Classification encoder-decoder

来源：评论

学校读者我要写书评

暂无评论

Multi-step ahead wind power forecasting based on dual-attention mechanism

引用

ENERGY REPORTS 2023年 9卷 239-251页

作者： Aslam, Muhammad Kim, Jun-Sung Jung, Jaesung Sejong Univ Dept Artificial Intelligence Seoul 05006 South Korea Korea Elect Power Res Inst Digital Solut Lab Daejeon 34056 South Korea Ajou Univ Dept Elect & Comp Engn Suwon 16499 Gyeonggi Do South Korea

Accurate forecasting is essential for the economic benefits and efficient operation of intermittent wind power systems. Multi-step ahead wind power forecasting provides multiple benefits in the planning and operation of the power systems. This study proposes a deep learning model based on a dual -attention mechanism for multi-step ahead wind power forecasting. Both the attention mechanisms are applied over the encoder-decoder based sequence-to-sequence model consisting of long short-term memory (LSTM) blocks. The Bayesian optimization algorithm is applied to the proposed model to obtain the optimal combination of hyper-parameters. To evaluate its effectiveness, the proposed model has been compared with the persistence model, and state-of-the-art models such as simple LSTM, LSTM-attention, ensemble model, and neural basis expansion analysis for time series (N-BEATS) model. Furthermore, the performance of the attention mechanism with respect to impactful input features, such as wind speed and air pressure, was analyzed. The forecasting skill score of the proposed model was the highest among all other models in comparison, which indicates the effectiveness of the proposed model. Similarly, the proposed model outperformed the traditional methods in terms of other evaluating criteria, including mean absolute error (MAE), and root mean square error (RMSE) among others, hence, proving its efficacy. The average RMSE score of the proposed model for multi horizon forecasting was 0.04995, whereas that of N-BEATS was 0.0876, ensemble method was 0.1132 and LSTM-attention was around 0.101375. Similarly, the average forecasting skill score of proposed method was 0.6625, whereas that of N-BEATS was 0.3975 and LSTM-attention achieved 0.3775 skill score. (c) 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://***/licenses/by-nc-nd/4.0/).

关键词： Attention mechanism Bayesian optimization encoder-decoder LSTM Renewable energy Wind power forecasting

来源：评论

学校读者我要写书评

暂无评论

A multi-scale inputs and labels model for background subtraction

引用

SIGNAL IMAGE AND VIDEO PROCESSING 2023年第8期17卷 4133-4141页

作者： Yang, Yizhong Li, Dajin Li, Xiang Zhang, Zhang Xie, Guangjun Hefei Univ Technol Sch Microelect Hefei Peoples R China

Background subtraction is a challenging and fundamental task in computer vision, which aims at segmenting moving objects from the background. Recently, the attention mechanism has become a hot topic in the neural network. The algorithms based on encoder-decoder and multi-scale type network perform impressive results in the domain of background subtraction. In this paper, we propose a multi-scale inputs and labels (MSIL) model which is based on the encoder-decoder type network and the channel attention. The multi-scale fusion encoding (MSFE) module aims to utilize multi-scale inputs effectively, which can fuse the high-level and low-level features details. The channel attention (CA) module is introduced to connect the encoder and decoder to model channel-wise attentions. The multi-label supervision decoding (MLSD) module helps to learn richer hierarchical features and achieves better performance by the new multi-label supervision. The proposed model is also evaluated on the CDnet-2014 dataset and the LASIESTA dataset, which demonstrate the effectiveness and superiority of the proposed model by an average F-Measure of 0.9851 and 0.9633, respectively. In addition, scene independent evaluation experiments on the CDnet-2014 dataset demonstrate the effectiveness of the model on unseen videos.

关键词： Background subtraction Neural network Multi-scale inputs and labels encoder-decoder

来源：评论

学校读者我要写书评

暂无评论

Designing an adaptive cost function for dynamic human pose predictions

引用

MULTIMEDIA TOOLS AND APPLICATIONS 2023年第18期83卷 53201-53219页

作者： Yadav, Gaurav Kumar Puig, Domenec Nandi, G. C. Indian Inst Informat Technol Allahabad Dept Informat Technol Prayagraj Uttar Pradesh India Univ Rovira I Virgili Dept Comp Sci & Math Secur Tarragona Spain

In the modern-day scenario, machines and humans are expected to work together and collaborate in several social and manufacturing environments. The machines should predict humans' next move for effective collaborations by observing their present move. Human motion modelling and prediction are fundamental and challenging problems involving computer vision and graphics. To help solve some of the challenges, in the present investigation, we propose an innovative idea of developing a new cost function as the objective function based on adaptive sampling, which is subsequently used with an 'Adam' optimizer for training and validating a specially configured Deep Learning architecture. Our proposed development produced significantly improved results regarding future pose estimation/predictions. The adaptiveness of the proposed cost function is based on a bell-shaped locally weighted function. It has been observed that the area covered by the cost function plays a vital role during training, and the bell-shaped function's width helps decide the region of importance for the training samples. The proposed cost function has been used for training a gated recurrent unit (GRU) based encoder-decoder architecture. The encoder takes the observed input sequences, extracts the input sequence's significant variability, and passes it to the decoder. The decoder takes it as input, trains using the adaptive sampling-based method, and predicts future poses. We have experimented with this function in various sizes and shapes and compared the results obtained with some state-of-the-art research results. As elaborated in this paper, we obtained much-improved results in almost all the cases.

关键词： Deep sequential networks encoder-decoder Human 3.6M

来源：评论

学校读者我要写书评

暂无评论

End-to-End Handwritten Paragraph Text Recognition Using a Vertical Attention Network

引用

IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE 2023年第1期45卷 508-524页

作者： Coquenet, Denis Chatelain, Clement Paquet, Thierry LITIS EA 4108 F-76800 Saint Etienne Du Rouvray France Univ Rouen Normandy F-76000 Rouen France Normandy Univ F-14032 Caen France INSA Rouen Normandy F-76800 Saint Etienne Du Rouvray France

Unconstrained handwritten text recognition remains challenging for computer vision systems. Paragraph text recognition is traditionally achieved by two models: the first one for line segmentation and the second one for text line recognition. We propose a unified end-to-end model using hybrid attention to tackle this task. This model is designed to iteratively process a paragraph image line by line. It can be split into three modules. An encoder generates feature maps from the whole paragraph image. Then, an attention module recurrently generates a vertical weighted mask enabling to focus on the current text line features. This way, it performs a kind of implicit line segmentation. For each text line features, a decoder module recognizes the character sequence associated, leading to the recognition of a whole paragraph. We achieve state-of-the-art character error rate at paragraph level on three popular datasets: 1.91% for RIMES, 4.45% for IAM and 3.59% for READ 2016. Our code and trained model weights are available at https://***/FactoDeepLearning/VerticalAttentionOCR.

关键词： Seq2Seq model hybrid attention segmentation-free paragraph handwriting recognition fully convolutional network encoder-decoder optical character recognition

来源：评论

学校读者我要写书评

暂无评论

Dual-View Spectral and Global Spatial Feature Fusion Network for Hyperspectral Image Classification

引用

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING 2023年 61卷 1页

作者： Guo, Tan Wang, Ruizhi Luo, Fulin Gong, Xiuwen Zhang, Lei Gao, Xinbo Chongqing Univ Posts & Telecommun Sch Commun & Informat Engn Chongqing 400065 Peoples R China Chongqing Univ Coll Comp Sci Chongqing 400044 Peoples R China Univ Sydney Fac Engn Camperdown NSW 2006 Australia Chongqing Univ Sch Microelect & Commun Engn Chongqing 400044 Peoples R China Chongqing Univ Posts & Telecommun Chongqing Key Lab Image Cognit Chongqing 400065 Peoples R China

For hyperspectral image (HSI) classification, two branch networks generally use convolutional neural networks (CNNs) to extract the spatial features and long short-term memory (LSTM) to learn the spectral features. However, CNNs with a local kernel neglect the global properties of the whole HSI. LSTM does not consider the macroscopic and detailed information of spectra. In this article, we propose a dual-view spectral and global spatial feature fusion network (DSGSF) to extract the spatial-spectral features for HSI classification (HSIC), including a spatial subnetwork and a spectral subnetwork. In the spatial subnetwork, we propose a global spatial feature representation model based on the encoder-decoder structure with channel attention and spatial attention to learn the global spatial features. In the spectral subnetwork, we design a dual-view spectral feature aggregation model with view attention to learn the diversity of spectral features. By fusing the two subnetworks, we construct DSGSF to extract the spatial-spectral features of HSI with strong discriminating performance. Experimental results on three public datasets illustrate that the proposed method can achieve competitive results compared with the state-of-the-art methods. Code: https://***/RZWang-WH/DSGSF.

关键词： Feature extraction Convolution Convolutional neural networks Data mining Semantics Radio frequency Kernel Attention encoder-decoder global feature hyperspectral image (HSI) classification long short view feature

来源：评论

学校读者我要写书评

暂无评论

Research on neural processes with multiple latent variables

引用

IET IMAGE PROCESSING 2023年第11期17卷 3323-3336页

作者： Yu, Xiao-Han Mao, Shao-Chen Wang, Lei Lu, Shi-Jie Yu, Kun Army Engn Univ PLA Coll Command & Control Syst Nanjing Peoples R China Army Engn Univ PLA Coll Commun Engineer Nanjing Peoples R China Army Engn Univ PLA Coll Command & Control Syst Nanjing 210007 Peoples R China

Neural Process (NP) fully combines the advantages of neural network and Gaussian Process (GP) to provide an efficient method for solving regression problems. Nonetheless, limited by the dimensionality of the latent variable, NP has difficulty fitting the observed data completely and predicting the targets perfectly. To remedy these drawbacks, the authors propose a concise and effective improvement of the latent path of NP, which the authors term Multi-Latent Variables Neural Process (MLNP). MLNP samples multiple latent variables and integrates the representations corresponding to the latent variables in the decoder with adaptive weights. MLNP inherits the desirable property of linear computation scales of NP and learns the approximate distribution over objective functions from contexts more flexibly and accurately. By applying MLNP to 1-D regression, real-world image completion, which can be seen as a 2-D regression task, the authors demonstrate its significant improvement in the accuracy of prediction and contexts fitting capability compared with NP. Through ablation experiments, the authors also verify that the number of latent variables has a great impact on the prediction accuracy and fitting capability of MLNP. Moreover, the authors also analyze the roles played by different latent variables in reconstructing images.

关键词： encoder-decoder multiple latent variables neural process regression

来源：评论

学校读者我要写书评

暂无评论

SATFace: Subject Agnostic Talking Face Generation with Natural Head Movement

引用

NEURAL PROCESSING LETTERS 2023年第6期55卷 7529-7542页

作者： Yang, Shuai Qiao, Kai Shi, Shuhao Yang, Jie Ma, Dekui Hu, Guoen Yan, Bin Chen, Jian PLA Strategy Support Force Informat Engn Univ Henan Key Lab Imaging & Intelligence Proc Sci Ave Zhengzhou 450001 Henan Peoples R China

Talking face generation is widely used in education, entertainment, shopping, and other social practices. Existing methods focus on matching the speaker's mouth shape with the speech content. Still, there is a lack of research on automatically extracting potential head motion features from speech, resulting in a lack of naturalness. This paper proposes SATFace, a subject agnostic talking face generation method with natural head movement. To model the talking face's complicated and critical features (identity, background, mouth shape, head posture, etc.), we construct SATFace by taking encoder-decoder as the primary network architecture. Then, we design a long short-time feature learning network to better reference the global and local information in audio for generating reasonable head movement. Besides, a modular training process is proposed to improve explicit and implicit features' learning effects and efficiency. The experimental comparison results show that SATFace improves by at least about 9.8% in cumulative probability of blur detection and 8.2% in synchronization confidence compared with the mainstream methods. The mean opinion scores show that SATFace has advantages in terms of lip sync quality, head movement naturalness, and video realness.

关键词： Talking face generation Generative adversarial networks encoder-decoder Feature learning

来源：评论

学校读者我要写书评

暂无评论

An abstractive text summarization technique using transformer model with self-attention mechanism

引用

NEURAL COMPUTING & APPLICATIONS 2023年第25期35卷 18603-18622页

作者： Kumar, Sandeep Solanki, Arun Gautam Buddha Univ Dept Comp Sci & Engn Greater Noida 201312 Uttar Pradesh India

Creating a summarized version of a text document that still conveys precise meaning is an incredibly complex endeavor in natural language processing (NLP). Abstract text summarization (ATS) is the process of using facts from source sentences and merging them into concise representations while maintaining the content and intent of the text. Manually summarizing large amounts of text are challenging and time-consuming for humans. Therefore, text summarization has become an exciting research focus in NLP. This research paper proposed an ATS model using a Transformer Technique with Self-Attention Mechanism (T2SAM). The self-attention mechanism is added to the transformer to solve the problem of coreference in text. This makes the system to understand the text better. The proposed T2SAM model improves the performance of text summarization. It is trained on the Inshorts News dataset combined with the DUC-2004 shared tasks dataset. The performance of the proposed model has been evaluated using the ROUGE metrics, and it has been shown to outperform the existing state-of-the-art baseline models. The proposed model gives the training loss minimum to 1.8220 from 10.3058 (at the starting point) up to 30 epochs, and it achieved model accuracy 48.50% F1-Score on both the Inshorts and DUC-2004 news datasets.

关键词： Abstractive text summarization encoder-decoder Self-attention mechanism Rouge metrics Transformer architecture T2SAM

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：