检索结果-内蒙古大学图书馆

MMInstruct: a high-quality multi-modal instruction tuning dataset with extensive diversity

science China(Information sciences) 2024年第12期67卷 36-51页

作者： Yangzhou LIU Yue CAO Zhangwei GAO Weiyun WANG Zhe CHEN Wenhai WANG Hao TIAN Lewei LU Xizhou ZHU Tong LU Yu QIAO Jifeng DAI School of Computer Science Nanjing University School of Electronic Information and Electrical Engineering Shanghai Jiao Tong University Shanghai AI Laboratory School of Computer Science Fudan University Department of Information Engineering The Chinese University of Hong Kong SenseTime Research Department of Electronic Engineering Tsinghua University

Despite the effectiveness of vision-language supervised fine-tuning in enhancing the performance of vision large language models(VLLMs), existing visual instruction tuning datasets include the following limitations.(1) Instruction annotation quality: despite existing VLLMs exhibiting strong performance,instructions generated by those advanced VLLMs may still suffer from inaccuracies, such as hallucinations.(2) Instructions and image diversity: the limited range of instruction types and the lack of diversity in image data may impact the model's ability to generate diversified and closer to real-world scenarios outputs. To address these challenges, we construct a high-quality, diverse visual instruction tuning dataset MMInstruct,which consists of 973k instructions from 24 domains. There are four instruction types: judgment, multiplechoice, long visual question answering, and short visual question answering. To construct MMInstruct, we propose an instruction generation data engine that leverages GPT-4V, GPT-3.5, and manual correction. Our instruction generation engine enables semi-automatic, low-cost, and multi-domain instruction generation at 1/6 the cost of manual construction. Through extensive experiment validation and ablation experiments,we demonstrate that MMInstruct could significantly improve the performance of VLLMs, e.g., the model fine-tuning on MMInstruct achieves new state-of-the-art performance on 10 out of 12 benchmarks. The code and data shall be available at https://***/yuecao0119/MMInstruct.

关键词： instruction tuning multi-modal multi-domain dataset vision large language model

来源：评论

学校读者我要写书评

暂无评论

A new method for promoting the detection of breast cancer in thermograms: applying deep autoencoders for eliminating redundancies in parallel with preserving independent components

引用

Journal of Ambient Intelligence and Humanized Computing 2024年第12期15卷 4085-4099页

作者： Shojaedini, Seyed Vahab Bahramzadeh, Bahram Biomedical Engineering Department Iranian Research Organization for Science & ampTechnology Tehran Iran Department of Computer Engineering Faculty of Engineering Islamic Azad University E-Campus Tehran Iran

Although thermography has been proposed over the past decade as an effective method for breast cancer diagnosis, the complexity of thermograms presents a significant obstacle, making their interpretation challenging. Recently, deep learning has emerged as a promising framework for analyzing breast thermograms. However, the performance of deep neural networks heavily relies on the volume of training images, and the limited number of labeled thermograms necessitates the generation of synthetic images for training. This article introduces a novel method that eliminates redundant or destructive information in synthetic thermograms while preserving essential and independent features using deep autoencoder concepts. Consequently, the proposed strategy enhances the representation of synthetic images for training deep networks, leading to improved breast cancer detection in thermograms. Evaluating the performance of the proposed method on the DMR-IR dataset shows a significant improvement in detecting cancerous breasts in thermograms compared to benchmark methods. Integrating this innovation into the basic model achieved an average accuracy of 92.3%, sensitivity of 93%, and specificity of 91.4%, surpassing the basic model’s 89.1%, 86%, and 92.5%, respectively. The proposed method demonstrated superior performance in mitigating overfitting, as evidenced by a smaller gap between training and validation curves, resulting in a 7% improvement in accuracy and a 3.2% increase in sensitivity. Although there was a slight decrease of 1.1% in specificity, it was outweighed by other parameter improvements. Additionally, the method led to a significant reduction in the dispersion range of test results, approximately six times less than alternative approaches. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.

关键词： Lung cancer

来源：评论

学校读者我要写书评

暂无评论

Learning continuous network emerging dynamics from scarce observations via data-adaptive stochastic processes

引用

science China(Information sciences) 2024年第12期67卷 240-255页

作者： Jiaxu CUI Qipeng WANG Bingyi SUN Jiming LIU Bo YANG College of Computer Science and Technology Jilin University Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University Public Computer Education and Research Center Jilin University Department of Computer Science Hong Kong Baptist University

Learning network dynamics from the empirical structure and spatio-temporal observation data is crucial to revealing the interaction mechanisms of complex networks in a wide range of domains. However,most existing methods only aim at learning network dynamic behaviors generated by a specific ordinary differential equation instance, resulting in ineffectiveness for new ones, and generally require dense *** observed data, especially from network emerging dynamics, are usually difficult to obtain, which brings trouble to model learning. Therefore, learning accurate network dynamics with sparse, irregularly-sampled,partial, and noisy observations remains a fundamental challenge. We introduce a new concept of the stochastic skeleton and its neural implementation, i.e., neural ODE processes for network dynamics(NDP4ND), a new class of stochastic processes governed by stochastic data-adaptive network dynamics, to overcome the challenge and learn continuous network dynamics from scarce observations. Intensive experiments conducted on various network dynamics in ecological population evolution, phototaxis movement, brain activity, epidemic spreading, and real-world empirical systems, demonstrate that the proposed method has excellent data adaptability and computational efficiency, and can adapt to unseen network emerging dynamics, producing accurate interpolation and extrapolation with reducing the ratio of required observation data to only about 6% and improving the learning speed for new dynamics by three orders of magnitude.

关键词： complex networks network dynamics emerging spatio-temporal dynamics neural processes

来源：评论

学校读者我要写书评

暂无评论

How far are we to GPT-4V?Closing the gap to commercial multimodal models with open-source suites

引用

science China(Information sciences) 2024年第12期67卷 5-22页

作者： Zhe CHEN Weiyun WANG Hao TIAN Shenglong YE Zhangwei GAO Erfei CUI Wenwen TONG Kongzhi HU Jiapeng LUO Zheng MA Ji MA Jiaqi WANG Xiaoyi DONG Hang YAN Hewei GUO Conghui HE Botian SHI Zhenjiang JIN Chao XU Bin WANG Xingjian WEI Wei LI Wenjian ZHANG Bo ZHANG Pinlong CAI Licheng WEN Xiangchao YAN Min DOU Lewei LU Xizhou ZHU Tong LU Dahua LIN Yu QIAO Jifeng DAI Wenhai WANG State Key Laboratory for Novel Software Technology Nanjing University Shanghai AI Laboratory School of Computer Science Fudan University SenseTime Research Department of Information Engineering The Chinese University of Hong Kong Department of Electronic Engineering Tsinghua University

In this paper, we introduce InternVL 1.5, an open-source multimodal large language model(MLLM) to bridge the capability gap between open-source and proprietary commercial models in multimodal understanding. We introduce three simple improvements.(1) Strong vision encoder: we explored a continuous learning strategy for the large-scale vision foundation model — InternViT-6B, boosting its visual understanding capabilities, and making it can be transferred and reused in different LLMs.(2) Dynamic high-resolution: we divide images into tiles ranging from 1 to 40 of 448×448 pixels according to the aspect ratio and resolution of the input images, which supports up to 4K resolution input.(3) High-quality bilingual dataset: we carefully collected a high-quality bilingual dataset that covers common scenes, document images,and annotated them with English and Chinese question-answer pairs, significantly enhancing performance in optical character recognition(OCR) and Chinese-related tasks. We evaluate InternVL 1.5 through a series of benchmarks and comparative studies. Compared to both open-source and proprietary commercial models, InternVL 1.5 shows competitive performance, achieving state-of-the-art results in 8 of 18 multimodal benchmarks. Code and models are available at https://***/OpenGVLab/InternVL.

关键词： multimodal model open-source vision encoder dynamic resolution bilingual dataset

来源：评论

学校读者我要写书评

暂无评论

Bioinspired design of integral molded Janus silk fibroin-MXene evaporator for efficient solar vapor generation

引用

Nano research 2024年第4期17卷 2824-2835页

作者： Ling-Qing Kong Zhi-Cheng Zeng Guang-Zong Min Zhao-Hui Meng Guo-Qing Meng Xiao-Xiao Guo Nai-Bo Lin Xiang-Yang Liu The Higher Educational Key Laboratory for Biomedical Engineering of Fujian Province Research Center of Biomedical Engineering of XiamenDepartment of BiomaterialsCollege of MaterialsXiamen UniversityXiamen 361005China The State Key Laboratory of Marine Environmental Science(MEL) College of Ocean and Earth SciencesXiamen UniversityXiamen 361005China

Solar vapor generation is a promising sustainable technology that uses solar distillation to produce fresh water from seawater and wastewater,helping relieve global water resource ***,inspired by naturally grown integrally molded mulberry leaves with a Janus hydrophilic and hydrophobic structure,a novel,simple,and efficient integrated molding method is proposed to break through the limitations of the traditional split manufacturing strategy and realizes the integrated formation of Janus *** on the spontaneous sedimentation characteristics of MXene in silk fibroin solution and its regulation of mesoscopic structure and hydrophilicity of silk fibroin,layered structures with different compositions and hydrophilicities were obtained in one ***,ethanol and glutaraldehyde were added to construct a physical crystallization-chemical crosslinking dual stabilization structure in silk *** evaporator has the evaporation rate of 3.07 kg·m^(-2)·h^(-1) and the efficiency of 86.8%under 1 sun and maintains high evaporation performance under various extreme test conditions including vigorous washing,repeated compression,and high-intensity ultraviolet(UV)***,the evaporator performs well in practical application scenarios,its evaporation rate in the simulated Dead Sea seawater exceeds 2.13 kg·m^(-2)·h^(-1),and more than 99.9%of the salt,heavy metal ions,oil pollution,and dyes are purified.

关键词： silk fibroin MXene bioinspired integrated molding Janus structure solar vapor generation

来源：评论

学校读者我要写书评

暂无评论

Mixed-decomposed convolutional network:A lightweight yet efficient convolutional neural network for ocular disease recognition

引用

CAAI Transactions on Intelligence Technology 2024年第2期9卷 319-332页

作者： Xiaoqing Zhang Xiao Wu Zunjie Xiao Lingxi Hu Zhongxi Qiu Qingyang Sun Risa Higashita Jiang Liu Research Institute of Trustworthy Autonomous Systems and Department of Computer Science and Engineering Southern University of Science and TechnologyShenzhenChina Tomey Corporation NagoyaJapan Guangdong Provincial Key Laboratory of Brain‐inspired Intelligent Computation Department of Computer Science and EngineeringSouthern University of Science and TechnologyShenzhenChina Singapore Eye Research Institute SingaporeSingapore

Eye health has become a global health concern and attracted broad *** the years,researchers have proposed many state-of-the-art convolutional neural networks(CNNs)to assist ophthalmologists in diagnosing ocular diseases efficiently and ***,most existing methods were dedicated to constructing sophisticated CNNs,inevitably ignoring the trade-off between performance and model *** alleviate this paradox,this paper proposes a lightweight yet efficient network architecture,mixeddecomposed convolutional network(MDNet),to recognise ocular *** MDNet,we introduce a novel mixed-decomposed depthwise convolution method,which takes advantage of depthwise convolution and depthwise dilated convolution operations to capture low-resolution and high-resolution patterns by using fewer computations and fewer *** conduct extensive experiments on the clinical anterior segment optical coherence tomography(AS-OCT),LAG,University of California San Diego,and CIFAR-100 *** results show our MDNet achieves a better trade-off between the performance and model complexity than efficient CNNs including MobileNets and ***,our MDNet outperforms MobileNets by 2.5%of accuracy by using 22%fewer parameters and 30%fewer computations on the AS-OCT dataset.

关键词： artificial intelligence deep learning deep neural networks image analysis image classification medical applications medical image processing

来源：评论

学校读者我要写书评

暂无评论

MMP-9-responsive probe for fiuorescence-magnetic resonance dual-mode imaging of hepatocellular carcinoma models with different metastatic capacities

引用

Chinese Chemical Letters 2025年第04期36卷 381-386页

作者： Qiuye Wang Yabing Sun Liangxue Lai Haijing Cui Yonglong Ye Ming Yang Weihao Zhu Bo Yuan Quanliang Mao Wenzhi Ren Aiguo Wu Ningbo Key Laboratory of Biomedical Imaging Probe Materials and Technology Zhejiang International Cooperation Base of Biomedical Materials Technology and Application Chinese Academy of Sciences (CAS) Key Laboratory of Magnetic Materials and Devices Ningbo Cixi Institute of Biomedical Engineering Zhejiang Engineering Research Center for Biomedical Materials Ningbo Institute of Materials Technology and Engineering Chinese Academy of Sciences Cixi Biomedical Research Institute Wenzhou Medical University Advanced Energy Science and Technology Guangdong Laboratory Department of Radiology First Affiliated Hospital of Ningbo University

Visual assessment of tumor metastatic capacity is crucial for predicting hepatocellular carcinoma(HCC)prognosis and guiding clinical therapeutic approaches. In this study, we developed an enzyme-responsive probe based on the peptide GK10, which is selectively cleaved by matrix metalloproteinase-9(MMP-9),a critical marker for metastasis in HCC. The GK10 peptide was conjugated with near-infrared fiuorescent molecule IR783, fiuorescent quencher black hole quencher 3(BHQ3), and magnetic resonance(MR) contrast agent DOTA-Gd, forming the IR783-GK10-BHQ3-Gd probe. Upon MMP-9 cleavage of GK10, BHQ3 is released from the probe, thereby amplifying the previously quenched IR783 fiuorescence signal. In vitro experiments demonstrate the probe's impressive detection limit for MMP-9, as low as 1.84 ng/m L. Moreover, in vivo imaging results reveal that the probe can differentiate liver cancers with varying metastatic capacities. The fiuorescence and MR imaging signal intensity of high metastatic HCC are approximately1.2 times greater than that of low metastatic HCC. Thus, this engineered probe holds promise as a valuable tool for evaluating HCC metastatic capacity through fiuorescence-MR dual-mode imaging.

关键词： Fluorescence imaging Förster resonance energy transfer Hepatocellular carcinoma Magneticresonance imaging Matrix metalloproteinase-9

来源：评论

学校读者我要写书评

暂无评论

Green hydrogen production from photovoltaic power station as a road map to climate change mitigation

引用

Clean Energy 2024年第2期8卷 156-167页

作者： Kehinde Ridwan Kamil Bassey Okon Samuel Umar Khan Department of Computer Science and Engineering Edge Hill UniversityUK Enel Green Power Rome Italy Materials Research Laboratory Steelagon Engineering LimitedKaduna StateNigeria

The increasing recognition of hydrogen as a critical element in the global net-zero transition and its clear role in decarbonizing challenging sectors coincide with the growing urgency to address climate ***'s favourable renewable-energy capacity,ranging from 28%to 36%for solar,has been reported by the global solar irradiance ***,the majority of hydrogen production today relies on fossil fuels(96%),with only a small fraction(4%)being produced through water *** though there have been many studies on climate change mitigation with a focus on Africa,a green hydrogen production from a photovoltaic power station approach has not been ***,literature with a focus on Nigeria is *** study focuses on the African green hydrogen production industry,utilizing Nigeria as a case study to explore the feasibility of generating clean hydrogen vectors from a percentage of photovoltaic power output in various regions of the country through stand-alone solar grid electrification *** of the usage and effectiveness of the produced hydrogen fuel in each region are carried out,with the highest region having an annual output of 12247278 kg of green hydrogen and 8573094 kg of ammonia and the lowest region having an output of 511245 kg of green hydrogen and 357871 kg of ammonia,and the expected production from the proposed usage of 50%of the power generation output of the installed 1.6-MWp and 80-kWp solar power minigrids in the regions is *** analyses were repeated for the other considered regions in the *** results showcased the enormous advantages of the electrolytic production of hydrogen and how the greener economy project can play a major role in mitigating climate change effects and overreliance on fossil fuels as the driver of the economy in many African countries.

关键词： renewable energy photovoltaic climate change green hydrogen vector fossil fossil fuels decarbonization electrolysis

来源：评论

学校读者我要写书评

暂无评论

Tunning the bandgap of MnO_(2)homojunction by building active high-index facet to achieve rapid electron transfer for enhanced photocatalytic sterilization

引用

Journal of Materials science & Technology 2024年第1期168卷 265-275页

作者： Tao Zhang Bo Li Chaofeng Wang Shuilin Wu Shengli Zhu Hui Jiang Yufeng Zheng Zhaoyang Li Zhenduo Cui Yu Zhang Paul K Chu Xiangmei Liu Biomedical Materials Engineering Research Center Hubei Key Laboratory of Polymer MaterialsMinistry-of-Education Key Laboratory for the Green Preparation and Application of Functional MaterialsSchool of Materials Science and Engineering Hubei UniversityWuhan 430062China School of Health Science&Biomedical Engineering Hebei University of TechnologyTianjin 300401China School of Materials Science&Engineering Peking UniversityBeijing 100871China School of Materials Science&Engineering the Key Laboratory of Advanced Ceramics and Machining Technology by the Ministry of Education of ChinaTianjin UniversityTianjin 300072China Department of Orthopedics Guangdong Provincial People’s HospitalGuangdong Academy of Medical SciencesGuangzhou 510080China Department of Physics Department of Materials Science and Engineering and Department of Biomedical EngineeringCity University of Hong KongHong Kong 999077China

Photocatalysis has been a research hotspot in recent years,and the design and modification of photocat-alysts have been the key *** methods for designing photocatalysts,including constructing heterojunctions and homojunctions,have been developed on the basis of *** this study,two homojunctions of manganese dioxide(MnO_(2)),including a high-index crystal plane homojunction and a general homojunction,are prepared using a stepwise hydrothermal *** a capping agent,the high-index crystal surface of the MnO_(2)is *** is found that the electron transport efficiency be-tween the two components of the homojunction with high-index planes is higher and the adsorption capacity of the oxygen is stronger,which leads to higher photocatalytic *** addition,the newly designed high-index homojunction is used for the treatment of bacterial infections,and it kills Staphy-lococcus aureus(***)and Escherichia coli(***)at rates of 99.95%±0.04%and 99.31%±0.25%,*** also has excellent therapeutic effects on mouse wounds,which implies superb practical application *** work provides a new strategy for the improved design of homojunctions and the application of photocatalytic materials.

关键词： Photocatalysis Homojunction Bandgap High-index facet Antibacterial

来源：评论

学校读者我要写书评

暂无评论

Risk Factors Categorizations of Ischemic Heart Disease in South-Western Bangladesh

引用

Data Intelligence 2024年第3期6卷 834-868页

作者： M.Raihan Sami Azam Laboni Akter Md.Mehedi Hassan Ryana Quadir Asif Karim Saikat Mondal Arun More Department of Computer Science and Engineering North Western UniversityKhulna 9100Bangladesh Faculty of Science and Technology Charles Darwin UniversityNT 0909Australia Departmentof Biomedical Engineering Khulna University of Engineering&TechnologyKhulnaBangladeh Computer Science and Engineering Discipline Khulna UniversityKhulna 9208Bangladesh Department of Computer Science and Engineering Daffodil International University(DIU)Bangladesh Department of Cardiology Ter Institute of Rural Health and ResearchMurud-413510India

Ischemic heart disease(IHD)is one of the leading causes of death ***,different geographic regions show different variations of the risk factors of this disease based on the different lifestyles of *** study examines the current IHD condition in southern Bangladesh,a Southeast Asian middle-income *** main approach to this research is an Al-based proposal of a reduced set of the greatest impact clinical traits that may cause *** approach attempts to reduce IHD morbidity and mortality by early detection of risk factors using the reduced set of clinical ***,diagnostic,and symptomatic features were considered for analysing this clinical *** pre-processing utilizes several machine learning techniques to select significant features and make meaningful interpretations.A proposed voting mechanism ranked the selected 138 features by their impact *** this regard,diverse patterns in correlations with variables,including age,sex,career,family history,obesity,etc.,were calculated and explained in terms of voting *** the 138 risk factors,three labels were categorized:high-risk,medium-risk,and low-risk features;19 features were regarded as high,25 were medium,and 94 were considered low impactful *** research's technological methodology and practical goals provide an innovative and resilient framework for addressing IHD,especially in less developed cities and townships of Bangladesh,where the general population's socioeconomic conditions are often *** data collection,pre-processing,and use of this study's complete and comprehensive IHD patient dataset is another innovative *** believe that other relevant research initiatives will benefit from this work.

关键词： Ischemic heart disease Machine learning CVD Data categorization Medical data

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：