检索结果-内蒙古大学图书馆

IEEE International Conference on Acoustics, Speech and Signal Processing

作者： S. Takamichi T. Toda A. W. Black S. Nakamura Grad. Sch. of Inf. Sci. Nara Inst. of Sci. & Technol. (NAIST) Nara Japan

ISBN: (纸本)9781467369985

This paper presents a novel training algorithm for Gaussian Mixture Model (GMM)-based Voice Conversion (VC). One of the advantages of GMM-based VC is computationally efficient conversion processing enabling to achieve real-time VC applications. On the other hand, the quality of the converted speech is still significantly worse than that of natural speech. In order to address this problem while preserving the computationally efficient conversion processing, the proposed training method enables 1) to use a consistent optimization criterion between training and conversion and 2) to compensate a Modulation Spectrum (MS) of the converted parameter trajectory as a feature sensitively correlated with over-smoothing effects causing quality degradation of the converted speech. The experimental results demonstrate that the proposed algorithm yields significant improvements in term of both the converted speech quality and the conversion accuracy for speaker individuality compared to the basic training algorithm.

关键词： GMM-based voice conversion modulation spectrum over-smoothing training algorithm modulation spectra e-bank vital capacity measurement training methods optimal criterion Speech Voice trajectory Functional training

来源：评论

学校读者我要写书评

暂无评论

Overcoming Hardware Imperfections in Optical Neural Networks Through a Machine Learning-Driven Self-Correction Mechanism

IEEE PHOTONICS JOURNAL

引用

IEEE PHOTONICS JOURNAL 2024年第2期16卷 1页

作者： Kim, Minjoo Kim, Beomju Kim, Yelim Handriani, Lia Saptini Jang, Suhee Jeong, Dae Yeop Yang, Sung Ik Park, Won Il Hanyang Univ Div Mat Sci & Engn Seoul 04763 South Korea Kyung Hee Univ Dept Appl Chem Yongin 17104 South Korea

We developed an optical neural network (ONN) for efficient processing and recognition of 2-dimensional (2D) images, employing a conventional liquid crystal display panel as optical neurons and synapses. This configuration allowed for optical signal outputs proportional to matrix-vector multiplication for 2D image inputs. However, our experimental results revealed a 26.6% decrease in the optical classification accuracy, despite utilizing digitally pre-trained parameters with 100% accuracy for 500 handwritten digits. This decline can be attributed to system imperfections associated with non-ideal functions of optical components and optical alignment. Rather than pursuing an elusive, imperfection-free ONN or attempting to calibrate these defects individually, we addressed these challenges by introducing a self-correction mechanism that utilizes a machine learning algorithm. This approach effectively restored the recognition accuracy and minimized loss of our ONN to levels comparable to the digitally pre-trained model. This study underscores the potential of constructing defect-tolerant hardware in ONNs through the application of machine learning techniques.

关键词： Hardware imperfections matrix-vector multiplication (MVM) optical neural networks (ONNs) recognition accuracy self-correction approach training algorithm

来源：评论

学校读者我要写书评

暂无评论

Recent advancement of remaining useful life prediction of lithium-ion battery in electric vehicle applications: A review of modelling mechanisms, network configurations, factors, and outstanding issues

引用

ENERGY REPORTS 2024年 11卷 4824-4848页

作者： Reza, M. S. Mannan, M. Mansor, M. Ker, Pin Jern Mahlia, T. M. Indra Hannan, M. A. Univ Tenaga Nas Dept Elect & Elect Engn Kajang 43000 Malaysia Univ Technol Sydney Sch Civil & Environm Engn Ultimo NSW 2007 Australia Sunway Univ Sch Engn & Technol Bandar Sunway 47500 Malaysia Korea Univ Sch Elect Engn Seoul 136701 South Korea

The remaining useful life (RUL) prediction of lithium -ion batteries (LIBs) plays a crucial role in battery management, safety assurance, and the anticipation of maintenance needs for reliable electric vehicle (EV) operation. An efficient prediction of RUL can ensure its safe operation and prevent both internal and external failures, as well as avoid any unwanted catastrophic events. However, achieving precise RUL prediction for electric vehicles presents a challenging task due to several issues related to intricate operational characteristics and dynamic shifts in model parameters throughout the aging process, battery parameters data extraction, data preprocessing, and hyperparameters tuning of the prediction model. This phenomenon significantly impacts the advancement of electric vehicle technology. To address these challenges, this study offers a comprehensive overview of various RUL prediction methods, presenting a comparative analysis of their outcomes, advantages, drawbacks, and associated research constraints. Emphasis is placed on the necessity of a battery management system (BMS) to ensure the safe and reliable functioning of LIBs. The review delves into crucial implementation factors, including battery test bench considerations, data selection, feature extraction, data preprocessing, performance evaluation indicators, and hyperparameter tuning. Additionally, the issues and challenges related to RUL prediction approaches such as;thermal runaway, material selection, cell balancing, battery aging, relaxation impact, training algorithms, data acquisition, and hyperparameter tuning were outlined to provide an in-depth understanding of the recent situations. The outcome of this review comprehensively examines various methods for predicting the RUL of LIB in EV applications, offering insights into their advantages, limitations, and research challenges. Recommendations for future trends in LIBs technology comprise enhancing prognostic accuracy and developing robus

关键词： Lithium -ion battery Remaining useful life Electric vehicle Battery management system Prediction method Battery aging training algorithm

来源：评论

学校读者我要写书评

暂无评论

BASS: Broad Network Based on Localized Stochastic Sensitivity

引用

IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS 2024年第2期35卷 1681-1695页

作者： Wang, Ting Zhang, Mingyang Zhang, Jianjun Ng, Wing W. Y. Chen, C. L. Philip South China Univ Technol Sch Med Guangzhou Peoples Hosp 1 Dept Radiol Guangzhou 510006 Peoples R China South China Univ Technol Sch Comp Sci & Engn Guangdong Prov Key Lab Computat Intelligence & Cy Guangzhou 510006 Peoples R China Pazhou Lab Brain & Affect Cognit Res Ctr Guangzhou 510335 Peoples R China South China Univ Technol Sch Comp Sci & Engn Guangzhou 510006 Peoples R China Dalian Maritime Univ Nav Coll Dalian 116026 Peoples R China

The training of the standard broad learning system (BLS) concerns the optimization of its output weights via the minimization of both training mean square error (MSE) and a penalty term. However, it degrades the generalization capability and robustness of BLS when facing complex and noisy environments, especially when small perturbations or noise appear in input data. Therefore, this work proposes a broad network based on localized stochastic sensitivity (BASS) algorithm to tackle the issue of noise or input perturbations from a local perturbation perspective. The localized stochastic sensitivity (LSS) prompts an increase in the network's noise robustness by considering unseen samples located within a Q-neighborhood of training samples, which enhances the generalization capability of BASS with respect to noisy and perturbed data. Then, three incremental learning algorithms are derived to update BASS quickly when new samples arrive or the network is deemed to be expanded, without retraining the entire model. Due to the inherent superiorities of the LSS, extensive experimental results on 13 benchmark datasets show that BASS yields better accuracies on various regression and classification problems. For instance, BASS uses fewer parameters (12.6 million) to yield 1% higher Top-1 accuracy in comparison to AlexNet (60 million) on the large-scale ImageNet (ILSVRC2012) dataset.

关键词： Broad learning system (BLS) generalization error localized stochastic sensitivity (LSS) training algorithm

来源：评论

学校读者我要写书评

暂无评论

Neural networks trained by weight permutation are universal approximators

引用

NEURAL NETWORKS 2025年 187卷 107277页

作者： Cai, Yongqiang Chen, Gaohang Qiao, Zhonghua Beijing Normal Univ Sch Math Sci Lab Math & Complex Syst MOE Beijing 100875 Peoples R China Hong Kong Polytech Univ Dept Appl Math Hung Hom Kowloon Hong Kong Peoples R China Hong Kong Polytech Univ Inst Smart Energy Dept Appl Math & Res Hung HomKowloon Hong Kong Peoples R China

The universal approximation property is fundamental to the success of neural networks, and has traditionally been achieved by training networks without any constraints on their parameters. However, recent experimental research proposed a novel permutation-based training method, which exhibited a desired classification performance without modifying the exact weight values. In this paper, we provide a theoretical guarantee of this permutation training method by proving its ability to guide a ReLU network to approximate onedimensional continuous functions. Our numerical results further validate this method's efficiency in regression tasks with various initializations. The notable observations during weight permutation suggest that permutation training can provide an innovative tool for describing network learning behavior.

关键词： Universal approximation property Neural networks training algorithm Learning behavior

来源：评论

学校读者我要写书评

暂无评论

Radar Signal Recognition Based on CSRDNN Network

引用

IEEE ACCESS 2024年 12卷 86704-86715页

作者： Zhang, Zheng Wan, Chuan Chen, Yi Zhou, Fang Zhu, Xiaofei Zhai, Wenchao Quan, Daying China Jiliang Univ Sch Informat Engn Hangzhou 310018 Peoples R China Xian Res Inst High Technol Xian 710025 Peoples R China

It is essential to achieve the high-accuracy recognition of low probability of intercept (LPI) radar signals in modern electronic warfare. However, under low signal-to-noise ratio (SNR), the recognition accuracy of the LPI radar signals is relatively low. In this paper, a novel radar signal recognition method based on Convolutional Stacked Recurrent Deep Neural Network (CSRDNN) is proposed. Firstly, we design a Convolutional Neural Network (CNN) to expand the feature space of input time domain signals, the features extracted by CNN were then used as inputs of the Stacked Recurrent Neural Networks (SRNN) module. In the SRNN module, we sequentially stack GRU, LSTM, and BGRU, enabling the model to better handle the short-term and long-term dependence of signal features and effectively solve asynchronous problems in unidirectional RNN networks. Subsequently, a Fully Connected Deep Neural Network (FCDNN) was employed to accomplish the recognition task. In addition, we design a training algorithm composed of the Nesterov-Adaptive Moment Estimation (Nadam) algorithm and the CosineAnnealing Learning Rate (LR) adjustment strategy to improve the training efficiency of the model. The experimental results demonstrate that the proposed model has higher recognition accuracy at low SNR compared to other models, with an overall recognition accuracy of 92.96% at -4 dB.

关键词： Long short term memory training data Logic gates Recurrent neural networks Feature extraction Radar signal processing algorithm design and analysis Low probability of intercept radar signal recognition stacked recurrent neural networks activation function training algorithm

来源：评论

学校读者我要写书评

暂无评论

A novel neural network training framework with data assimilation

引用

JOURNAL OF SUPERCOMPUTING 2022年第17期78卷 19020-19045页

作者： Chen, Chong Dou, Yixuan Chen, Jie Xue, Yaru China Univ Petr Coll Informat Sci & Engn Beijing 102249 Peoples R China

In recent years, the prosperity of deep learning has revolutionized the Artificial Neural Networks. However, the dependence of gradients and the offline training mechanism in the learning algorithms prevents the Artificial Neural Networks from further improvement. In this study, a gradient-free training framework based on data assimilation is proposed to avoid the calculation of gradients. In data assimilation algorithms, the error covariance between the forecasts and observations is used to optimize the states. The Feedforward Neural Networks are trained by gradient decent, data assimilation algorithms (Ensemble Kalman Filter and Ensemble Smoother with Multiple Data Assimilation), respectively. Ensemble Smoother with Multiple Data Assimilation trains Feedforward Neural Networks with pre-defined iterations by updating the parameters (i.e. states) using all the available observations which can be regarded as offline learning. Ensemble Kalman Filter optimizes Feedforward Neural Networks when new observation available by updating parameters which can be regarded as real-time learning. Two synthetic cases with the regression of a Sine function and a Mexican Hat function are conducted to validate the effectiveness of the proposed framework. Quantitative comparison with the root mean square error and coefficient of determination show that better performance is obtained by the proposed framework than the gradient decent method. Furthermore, the uncertainty of the parameters is quantified which shows the reduction in uncertainty along with the iterations in Ensemble Smoother with Multiple Data Assimilation. The proposed framework explores alternatives for real-time/offline training the existing Artificial Neural Networks (e.g. Convolutional Neural Networks, Recurrent Neural Networks) without the dependence of gradients and conducting uncertainty analysis at the same time.

关键词： Neural networks training algorithm Data assimilation EnKF ESMDA

来源：评论

学校读者我要写书评

暂无评论

Metaheuristic multi-objective optimization with artificial neural networks surrogate modeling for optimal energy-economic performance for CSP technology

引用

ENERGY AND AI 2025年 20卷

作者： Allouhi, A. Amine, M. Benzakour Aoul, K. A. Tabet USMBA Ecole Super Technol Fes Route ImouzzerBP 2427 Fes Morocco Univ Chouaib Doukkali Fac Sci El Jadida El Jadida 24000 Morocco United Arab Emirates Univ Coll Engn Architectural Engn Dept POB 15551 Al Ain U Arab Emirates

Among CSP technologies, the linear Fresnel reflector (LFR) can provide reliable carbon-neutral electricity for large-scale applications. In this study, the performance of a large solar LFR power plant under varying climatic conditions and the dependency of the performance on major plant design specifications, such as solar multiple and full-load thermal storage hours, were examined. Next, artificial neural network (ANN) surrogate models were introduced to predict the annual capacity factor of 100 MWe power plants operating with LFR technology. Single-hidden-layer ANN models with different numbers of neurons in the hidden layer were used and the Levenberg-Marquardt training algorithm was adopted. To overcome overfitting, validation and Bayesian Regularization approaches were compared. As training and testing data, 36 geographical sites with various combinations of design parameters were used. Through multi-objective optimization techniques, including the Multi- Objective Particle-Swarm Optimizer and Multi-Objective Grey Wolf Optimizer coupled with ANN surrogate modeling, this study navigates the trade-offs to identify Pareto-optimal solutions for large-scale LFR-based CSP integration based on the energy and cost criteria. The study also identified Site 4 (S4) as a promising candidate for optimal balance between the capacity factor (51.05%) and specific cost (5246.71$/kW), showcasing the practical implications of the research for sustainable and efficient CSP plant implementation.

关键词： Artificial neural network Capacity factor Linear fresnel reflector Multi-objective optimization training algorithm Metaheuristics

来源：评论

学校读者我要写书评

暂无评论

A Multi-Objective Decision-Making Neural Network: Effective Structure and Learning Method

引用

CONCURRENCY AND COMPUTATION-PRACTICE & EXPERIENCE 2025年第4-5期37卷

作者： Yan, Shu-Rong Nadershahi, Mohadeseh Guo, Wei Ghaderpour, Ebrahim Mohammadzadeh, Ardashir Guangzhou Huashang Coll Sch Digital Finance Guangzhou Peoples R China Payame Noor Univ Dept Ind Engn Tehran Iran Guangdong Univ Finance Sch Credit Management Guangzhou Peoples R China Sapienza Univ Rome Dept Earth Sci Rome Italy Sakarya Univ Dept Elect & Elect Engn Sakarya Turkiye

Decision Neural Networks significantly improve the performance of complex models and create more transparent and accountable decision-making systems that can be trusted in critical applications. However, their performance strongly depends on the amount of data and the learning algorithm. This article describes the development of a simplified structure and training algorithm based on the Levenberg-Marquardt algorithm to enhance the decision neural network's training and assess the utility function's efficacy in multi-objective issues. The suggested algorithm converges faster than traditional algorithms. Also, the designed scheme combines gradient descent with the Gauss-Newton method, allowing it to escape shallow local minima more effectively than other similar techniques. Numerical examples demonstrate how well the suggested method estimates linear utility functions, even complicated and nonlinear ones. Additionally, the findings of applying the enhanced decision neural network to multi-objective decision-making issues show that this instructional technique produces responses with higher quality and faster convergence. By applying the designed scheme to a multi-objective problem with seven primary answers, it is shown that accuracy is improved by more than 20%.

关键词： decision neural network Levenberg-Marquardt algorithm multi-objective decision-making training algorithm utility function

来源：评论

学校读者我要写书评

暂无评论

Predicting total biogas potential of food waste using the initial output of biogas potential tests as input data to train an artificial neural network

引用

BIORESOURCE TECHNOLOGY REPORTS 2024年 26卷

作者： Hunter, Sarah M. Blanco, Edgar Borrion, Adiuan UCL Dept Civil Environm & Geomat Engn Chadwick BldgGower St London WC1E 6BT England Anaero Technol Ltd Cowley Rd Cambridge CB4 0DL England

Quantification of biogas potential is important for predicting anaerobic digestion operability and price. This study uses data from 446 biogas potential tests to train and test a multilayer perceptron artificial neural network (ANN) to forecast total biogas production using the evolution at the start of the experiment (3-14 days) as input data. ANN architecture (training algorithm, activation function, hidden nodes, regularisation, and input data) was optimised using response surface methodology. Best conditions (accuracy/computational speed) were obtained using adaptive moment estimation (adam) training algorithm and rectified linear unit (ReLU) activation function. When using three days of biogas production data, the accuracy of the model was reasonable (r2test = 0.881, r2validation = 0.879), although this increased significantly for 7 days (r2test = 0.953, r2validation = 0.925), or 14 days (r2test = 0.971, r2validation = 0.953). The highest accuracy was reported for readily digestible substrates (sugars and carbohydrates) and macronutrient mixtures. The methodology could be used to shorten prediction times of biogas potential tests.

关键词： Biogas yield Multilayer perceptron Response surface methodology Machine learning Activation function training algorithm

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：