检索结果-内蒙古大学图书馆

39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025

作者： Wang, Linqin Liu, Yaping Yu, Zhengtao Gao, Shengxiang Mao, Cunli Huang, Yuxin Wang, Wenjun Dong, Ling Faculty of Information Engineering and Automation Kunming University of Science and Technology Kunming China Yunnan Key Laboratory of Artificial Intelligence Kunming China

ISBN: (纸本)157735897X

With the rapid advancement of large language models (LLMs), discrete speech representations have become crucial for integrating speech into LLMs. Existing methods for speech representation discretization rely on a predefined codebook size and Euclidean distance-based quantization. However, 1) the size of codebook is a critical parameter that affects both codec performance and downstream task training efficiency. 2) The Euclidean distance-based quantization may lead to audio distortion when the size of the codebook is controlled within a reasonable range. In fact, in the field of information compression, structural information and entropy guidance are crucial, but previous methods have largely overlooked these factors. Therefore, we address the above issues from an information-theoretic perspective, we present SECodec, a novel speech representation codec based on structural entropy (SE) for building speech language models. Specifically, we first model speech as a graph, clustering the speech features nodes within the graph and extracting the corresponding codebook by hierarchically and disentangledly minimizing 2D SE. Then, to address the issue of audio distortion, we propose a new quantization method. This method still adheres to the 2D SE minimization principle, adaptively selecting the most suitable token corresponding to the cluster for each incoming original speech node. Furthermore, we develop a Structural Entropy-based Speech Language Model (SESLM) that leverages SECodec. Experimental results demonstrate that SECodec performs comparably to EnCodec in speech reconstruction, and SESLM surpasses VALL-E in zero-shot text-to-speech tasks. © 2025, Association for the Advancement of Artificial Intelligence (***). All rights reserved.

关键词： Encoding (symbols)

来源：评论

学校读者我要写书评

暂无评论

Research on Handwritten Marking Localization Method for Hot Rolled Steel Coils

Research on Handwritten Marking Localization Method for Hot ...

引用

Image Processing, Computer Vision and Machine Learning (ICICML), International Conference on

作者： Bowen Jia School of Information Engineering and Automation Kunming University of Science and Technology Kunming China

ISBN: (数字)9798350355413

ISBN: (纸本)9798350355420

Aiming at the problem that the handwritten markings of steel coils in the natural scenario of hot rolled steel coils warehouse in the travelling link using the existing text localization method have serious distortion and blurring, thus leading to a high leakage rate, a precise localization method for the handwritten markings of hot rolled steel coils is proposed. The method is based on Nanodet-m improved Coord-Nano steel coil handwritten labeling localization model. First, the backbone network perceives the spatial information of the feature map by introducing CoordConv and SAConc, which improves the adaptability and accuracy of the network in dealing with features of different scales and complexities. Then, the PfAAMLayer attention mechanism is introduced to produce a self-focusing effect on the features, which improves the architectural performance. Finally, the bottom-up path enhancement module is utilized to propagate the strong localization features in the shallow network to the deep network, so as to accurately locate the handwritten marking region of the steel roll from the complex environment. Experimental results show that the network model proposed in this paper achieves 97.7% localization accuracy for steel roll handwritten markers by training and validation on real steel roll handwritten markers image *** with the basic Nanodet-m, YOLOv7tiny and YOLOv8n, the localization accuracy of handwritten marks on steel rolls is improved by 1.6%, 0.8% and 0.2%.

关键词： Location awareness Coils Adaptation models Accuracy Attention mechanisms Object detection Distortion Steel Object recognition Labeling

来源：评论

学校读者我要写书评

暂无评论

Learning Ant Colony Algorithm for Solving Green Multi-Compartment Vehicle Routing Problem with Simultaneous Pickup and Delivery

Learning Ant Colony Algorithm for Solving Green Multi-Compar...

引用

IEEE International Conference on automation, Electronics and Electrical engineering (AUTEEE)

作者： Yan Geng Ning Guo Bin Qian Information Engineering and Automation Institute Kunming University of Science and Technology Kunming China

ISBN: (数字)9798350377033

ISBN: (纸本)9798350377040

To solve the Green Multi-Compartment Vehicle Routing Problem with Simultaneous Pickup and Delivery (GMCVRPSPD), a mathematical model is established with the objective of minimizing total costs. The model consists of vehicle operating costs and fuel consumption costs. A Learning Ant Colony Optimization algorithm (LACO) is proposed to solve GMCVRPSPD. Firstly, in the global search stage of LACO, three different heuristic methods are used to generate the initial solution. These solutions are used to initialize the pheromone matrix, providing a better starting point for the algorithm's subsequent search process. A pheromone update mechanism is designed to effectively enhance the convergence speed of the algorithm. Secondly, in the local search of LACO: a local learning strategy is designed, which selects the neighborhood operation suitable for the current search stage from the six neighborhood operations designed, and dynamically updates the selection probability value according to the contribution value generated by the neighborhood operation. In addition, a rapid evaluation strategy is designed to reduce the calculation time in the local search process. Finally, the effectiveness of LACO in solving GMCVRPSPD problem is verified by simulation experiment and algorithm comparison.

关键词： Integer programming Ant colony optimization Costs Heuristic algorithms Vehicle routing Transportation Search problems Mathematical models Optimization Convergence

来源：评论

学校读者我要写书评

暂无评论

A Mixed-Language Multi-Document News Summarization Dataset and a Graphs-Based Extract-Generate Model

arXiv

引用

arXiv 2024年

作者： Gao, Shengxiang Nan, Fang Zhang, Yongbing Huang, Yuxin Tan, Kaiwen Yu, Zhengtao Faculty of Information Engineering and Automation Kunming University of Science and Technology Kunming China Yunnan Key Laboratory of Artificial Intelligence Kunming University of Science and Technology Kunming China

Existing research on news summarization primarily focuses on single-language single-document (SLSD), single-language multi-document (SLMD) or cross-language single-document (CLSD). However, in real-world scenarios, news about a international event often involves multiple documents in different languages, i.e., mixed-language multi-document (MLMD). Therefore, summarizing MLMD news is of great significance. However, the lack of datasets for MLMD news summarization has constrained the development of research in this area. To fill this gap, we construct a mixed-language multi-document news summarization dataset (MLMD-news), which contains four different languages and 10,992 source document cluster and target summary pairs. Additionally, we propose a graph-based extract-generate model and benchmark various methods on the MLMD-news dataset and publicly release our dataset and code1, aiming to advance research in summarization within MLMD scenarios. Copyright © 2024, The Authors. All rights reserved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Profit-Driven Multi-Item Data Dissemination Based on Deep Reinforcement Learning in Internet of Vehicles

SSRN

引用

SSRN 2024年

作者： Wang, Shujuan Chen, Guangchao Li, Huafeng The School of Information Engineering and Automation Kunming University of Science and Technology Yunnan650500 China The School of Information Engineering and Automation Kunming University of Science and Technology Yunnan650500 China The School of Information Engineering and Automation Kunming University of Science and Technology Yunnan650500 China

Data dissemination plays a vital role in providing value-added services which enhance traffic safety and efficiency in the Internet of Vehicles (IoVs). However, major obstacles exist due to the limited communication resources and harsh transmission environment in IoVs. Moreover, existing work often assume the data/service providers offer their data/service for free, which weaken the practicability of these work. This paper focuses on the profit-driven multi-item data dissemination in IoVs. An incentive mechanism is established between RoadSide Units (RSUs) and vehicles, where vehicles are allowed to compete for receiving better service from RSUs, through a bidding strategy which dynamically adjusts a vehicle’s bidding price for a certain data item, based on multiple characteristics that affect the transmission performance. RSUs obtain the payments from vehicles as a compensation for the resource consumed in responding them. The efficient multi-item data dissemination problem is modeled as the optimization of RSU’s long-term revenue. Since the arrivals and distributions of vehicles’ requests, as well as the correlation between requested data items and the wireless channel conditions between RSUs and vehicles are usually unknown to the RSU a priori, RSU must learn the optimal scheduling policy that yields the greatest profit by interacting with the environment. A Double Deep Q Network (DDQN) based scheduling algorithm is proposed which implements intelligent data scheduling decisions. Meanwhile, the proposed algorithm overcomes the large-scale state problem and reduces Q value over-estimation. Simulation results show that the proposed strategy achieves significant performance improvement in terms of RSU’s revenue, successful response ratio and completion delay under varied environment, which justifies the effectiveness and efficiency of this work. © 2024, The Authors. All rights reserved.

关键词： Deep reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Interpretable prediction method of Matte grade based on GA-BiLSTM and SHAP

Interpretable prediction method of Matte grade based on GA-B...

引用

第43届中国控制会议

作者： Huaibo Ma Jun Ma Xiang Li Faculty of Information Engineering and Automation Kunming University of Science and Technology Yunnan Key Laboratory of Intelligent Control and Application Kunming University of Science and Technology

ISBN: (数字)9789887581581

ISBN: (纸本)9798350366907

Aiming at the problem that the ice-copper grade prediction process is difficult to effectively capture the long-term dependency relationship between data,which leads to low prediction accuracy,it is proposed to construct a model using the powerful long-term memory capability and bidirectional information prediction ability of bidirectional long and short-term memory network(BiLSTM) model,and further optimize the model by using the genetic algorithm(GA) to achieve accurate prediction of ice-copper ***,the genetic algorithm was used to find the optimal hyperparameter combination of the ***,the predicted values of the BiLSTM model optimized by the genetic algorithm are compared with the predicted values of the existing methods and the unoptimized Bi LSTM model using different evaluation ***,the SHAP method is used to give the overall and local interpretation of the prediction results of the BiLSTM *** experimental results show that the RMSE of the BiLSTM model optimized by the genetic algorithm is reduced by at least 3.05%,the MAE is reduced by at least 2.75%,the MAPE is reduced by at least 3.29%,and the R2 is increased by at least 1.51%,which verifies the effectiveness of the proposed method in solving the problem of ice-copper grade prediction,and the SHAP method is able to explain the output of the Bi LSTM model efficiently and increase the transparency of the *** SHAP method can effectively explain the output of the BiLSTM model and increase the transparency of the model.

关键词： Matte grade prediction Genetic algorithm BiLSTM model SHAP model interpretation

来源：评论

学校读者我要写书评

暂无评论

Efemnet: An Efficient Feature Extraction Multi-Attention Convolutional Neural Network for Skin Lesion Segmentation

SSRN

引用

SSRN 2024年

作者： Jing, Zijie Bai, Linlin Shao, Dangguo Ma, Lei Faculty of Information Engineering and Automation Kunming University of Science and Technology Kunming China School of Cyber Science and Engineering Zhengzhou University Zhengzhou China

Melanoma is the skin tumor with the highest mortality rate, and timely diagnosis based on dermoscopic images is an essential task in melanoma prevention and treatment. However, complex dermoscopic image morphology and unclear image edges affect the accurate diagnosis of melanoma. In this paper, we propose an encoder-decoder architecture (EFEMNet) to segment dermatologic lesions. First, we introduce residual concatenation in the encoder part to improve the feature retention ability;second, Coordinate Attention is utilized to identify and localize the target region more accurately for dermatoscopic images with different morphologies. Further, we propose an efficient feature extraction module (EFEM) to replace up-sampling operations to extract and fuse features efficiently. Finally, GAM Attention is added to the output layer to integrate the dimensions in space and channels to solve the problem of unclear edges in dermoscopy images. We evaluated the proposed method on the ISIC 2018 dataset and compared it with some state-of-the-art network models, EFEMNet segmented skin lesions more accurately, achieving 92.52% Dice, 86.81% IoU, 96.01% Accuracy, 93.86% Recall, and 92.92% Precision. Our method is superior to other methods in all evaluation indexes, and a series of ablation experiments also verify the effectiveness of the functional module. © 2024, The Authors. All rights reserved.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Anomaly Detection of Copper Melting Process Data Based on Anomaly Transformer

Anomaly Detection of Copper Melting Process Data Based on An...

引用

第43届中国控制会议

作者： Junjia Zhang Jun Ma Xiang Li Faculty of Information Engineering And Automation Kunming University of Science and Technology Yunnan Key Laboratory of Intelligent Control and Application Kunming University of Science and Technology

ISBN: (数字)9789887581581

ISBN: (纸本)9798350366907

In this paper,aiming at the copper melting process,the relevant process variable parameters are coupled with each other,complicated and change with the production cycle,which leads to the anomaly detection of copper melting *** accuracy of copper melting data,we propose an Anomaly Transformer model based on the Transformer model,which replaces the original Self-Attention mechanism with Anomaly-Attention mechanism,to realize accurate and reliable copper melting process data anomaly *** order to realize accurate and reliable anomaly detection of copper melting process ***,the degree of anomaly of copper melting process data at each time point is measured by the correlation dispersion,and the KL dispersion distance value is utilized to quantify the degree of *** value of KL dispersion to quantify the change of the value of each neighboring time point,and secondly,the above quantized distance value is transform the its attention mechanism into the anomaly attention mechanism form.,in order to present the anomaly deviation and weight of each time point,and distinguish normal points from anomaly points,and finally,the trained anomaly Transformer model is analyzed on the test set for anomaly ***,the trained anomaly Transformer model is analyzed on the test set for anomaly detection,and the results are analyzed and interpreted using F1 evaluation *** experimental results show that in the production process dataset of copper *** application of correlation-based discretization brings a certain 3.7%improvement in the F1 *** addition of the anomalous attention module achieved 43.4% performance improvement,and the final combination of the combination of the combined KL dispersion and the anomaly attention module yielded an F1 score of 71.31,achieving a better anomaly detection result.

关键词： Anomaly Detection of Copper Melting Process Data KL dispersion distance Anomaly-Attention Transformer

来源：评论

学校读者我要写书评

暂无评论

A Bayes CNN-LSTM approach for aero-engine remaining life predictions

A Bayes CNN-LSTM approach for aero-engine remaining life pre...

引用

第43届中国控制会议

作者： Jiangyan Zhu Jun Ma Jiande Wu Faculty of Information Engineering and Automation Kunming University of Science and Technology Yunnan Key Laboratory of Intelligent Control and Application Kunming University of Science and Technology

ISBN: (数字)9789887581581

ISBN: (纸本)9798350366907

Remaining life prediction(RUL) plays an important role in prognostic health *** recent years,deep learning-based methods have been widely used for RUL ***,deep learning-based methods all focus on point prediction of RUL,ignoring the uncertainty associated with inaccuracies in the data and *** the prominence of Bayesian methods in uncertainty modelling,this paper uses a Monte Carlo dropout(MC-Dropout)-based Bayesian approximation to capture the epistemic and contingent uncertainty in the prediction process.A Bayes CNN-LSTM-based spatial-temporal feature extraction model is proposed to implement RUL prediction and quantify *** proposed method is also validated on the CMAPSS dataset and compared with CNN,LSTM and CNN-LSTM to verify the effectiveness of Bayesian neural networks.

关键词： Remaining life prediction Bayesian neural networks Epistemic uncertainty Spatial-temporal feature

来源：评论

学校读者我要写书评

暂无评论

Hybrid Water Quality Prediction with Frequency Domain Conversion Enhancement and Seasonal Decomposition

Hybrid Water Quality Prediction with Frequency Domain Conver...

引用

2023 IEEE International Conference on Systems, Man, and Cybernetics, SMC 2023

作者： Bi, Jing Li, Yibo Chang, Xingyang Yuan, Haitao Qiao, Junfei Beijing University of Technology Faculty of Information Technology Beijing China School of Automation Science and Electrical Engineering Beihang University Beijing China

ISBN: (纸本)9798350337020

Water quality prediction can accurately reflect the development trend of water quality, and it is an important means to prevent the water environment from being polluted and maintain the health of the water environment. Existing prediction methods generally cannot accurately capture non-linear characteristics of water quality, and suffer from issues of gradient disappearance and gradient explosion. This work designs a water quality prediction model called SMF2 to effectively solve these problems and increase the accuracy of prediction. SMF2 combines the Savitsky-Golay filter, seasonal-trend decomposition using loess for multiple seasonal components, Fourier transform frequency-enhanced block and frequency-enhanced attention, serving for noise smoothing, extraction of exact seasonal components, time domain-frequency domain interconversion, feature extraction, and time series prediction by frequency domain low-rank approximation transform, respectively. Experimental results based on a real-life water environment data set show that the proposed SMF2 outperforms other advanced algorithms in terms of prediction accuracy. © 2023 IEEE.

关键词： Fourier transform frequency-enhanced block frequency-enhanced attention Savitsky-Golay filter seasonal-trend decomposition Water quality prediction

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：