[晓理紫]每日论文分享(有中文摘要，源码或项目地址)--大模型、扩散模型

专属领域论文订阅

VX 关注{晓理紫}，每日更新论文，如感兴趣，请转发给有需要的同学，谢谢支持

如果你感觉对你有所帮助，请关注我，每日准时为你推送最新论文。

分类:

大语言模型LLM
视觉模型VLM
扩散模型
视觉语言导航VLN
强化学习 RL
模仿学习 IL
机器人
开放词汇，检测分割

== LLM ==

标题: School evaluation and artificial intelligence

作者: Miguel Cobos, Henry Cherres

PubTime: 2023-12

Downlink: https://ieeexplore.ieee.org/document/10372877/

Journal: 2023 IEEE 3rd International Conference on Advanced Learning Technologies on Education & Research (ICALTER)

中文摘要: 教育中的评估随着时间的推移而发展，并建立了获取学生学业进步信息的新方法。然而，人工智能（如ChatGPT）的出现给评估过程带来了挑战，因为学生可以使用这些技术来解决问题和任务，而无需学习。这项研究的重点是推荐替代教育资源进行评估，考虑ChatGPT和其他人工智能的利弊。进行了系统的文献综述和笔试、Kahoot！、Quizlet、Mentimeter和Nearpod在Microsoft Excel中设计的工具中进行了识别和评估，以评估其有效性。结果显示，笔试和Plickers工具最有效，其次是ClassTools、Flip、Kahoot！、Quizlet、Mentimeter和Nearpod，此外，还显示了可用于通过量规评估知识的活动列表，其中最重要的有：口语课、展览、开放日、演讲、案例研究、辩论和参与观察，因为这些类型的工作鼓励演讲者准备和掌握主题。随着人工智能的出现，教育领域的传统评估面临着挑战，因为人工智能在学生不必读、写或学习的情况下生成任务和答案的能力对确保实际获得知识和技能提出了挑战。这项研究建议通过数字工具或教育活动来评估教育资源，这些工具或教育活动允许对学习进行真实的评估，并限制对人工智能的依赖。

摘要: Assessment in education has evolved over time and has established new ways of obtaining information about students’ academic progress. However, the advent of artificial intelligence, such as ChatGPT, has posed challenges in the assessment process, as students can use these technologies to solve questions and tasks without studying. This research focused on recommending alternative educational resources for assessment, considering the pros and cons of ChatGPT and other AI. A systematic literature review was conducted and resources such as written tests, Kahoot!, Quizlet, Mentimeter and Nearpod were identified and evaluated in the tool designed in Microsoft Excel to evaluate their effectiveness. The results showed that the written test and the Plickers tool were the most effective, followed by ClassTools, Flip, Kahoot!, Quizlet, Mentimeter and Nearpod, in addition a list of activities that can be used to assess knowledge through the rubric is shown, among the most important are: oral lessons, exhibitions, open houses, speeches, case studies, debates and observation of participation, because these types of work encourage the speaker to prepare and master the subject. Traditional assessment in the educational field has faced challenges with the advent of artificial intelligence because AI’s ability to generate tasks and answers without students having to read, write or study poses a challenge to ensure the actual acquisition of knowledge and skills. This research recommends educational resources to assess through digital tools or educational activities that allow for authentic assessment of learning and limit reliance on AI.

标题: Comparison of Classifiers for Text Classification in an E-commerce Service

作者: Çağdaş Doğan, Sinan Sarıca

PubTime: 2023-12

Downlink: https://ieeexplore.ieee.org/document/10415941/

Journal: 2023 14th International Conference on Electrical and Electronics Engineering (ELECO)

中文摘要: 在这项研究中，比较了几种文档矢量化技术和分类模型在文本分类方面的性能。这项工作是使用一个数据集完成的，该数据集由我们在物流行业提供的电子商务服务中收集的非结构化和手工标记的客户文本查询组成。这些查询是从通过书面沟通渠道提交给客户支持中心的原始文本中获得的。使用ChatGPT进行文本规范化，使用正则表达式在查询中标记实体，从而清理数据集。实验使用了三种矢量化方法，Tf-Idf，fastText和BERT，以及三种分类模型，梯度推进，支持向量机（SVM）和多层感知器（MLP）。每种矢量化方法与每种分类模型进行交叉验证，并对结果的准确性、召回率和F1分数进行评估。Tf-Idf方法在与MLP和SVM分类器一起使用时表现最佳。

摘要: In this study, performance of several document vectorization techniques and classification models are compared for text classification. The work was done with a dataset consisting of unstructured and hand-labeled customer text queries collected at the e-commerce service that we offer in the logistics industry. The queries were obtained from raw texts submitted to the customer support center through written communication channels. The dataset is cleaned up using ChatGPT for text normalization and regular expressions for tagging entities in queries. Three vectorization methods, Tf-Idf, fastText and BERT, and three classification models, gradient boosting, support vector machine (SVM) and multilayer perceptron (MLP), were used for the experiments. Each vectorization method was cross-validated with each classification model, and the results were evaluated for accuracy, recall, and F1 score. Tf-Idf method performed the best when used with MLP and SVM classifiers.

标题: Chat Generative Pretrained Transformer (ChatGPT) for Data Analysis

作者: Niranjanamurthy M, Kantharaju V, Sirisha Satish

PubTime: 2023-12

Downlink: https://ieeexplore.ieee.org/document/10435984/

Journal: 2023 3rd International Conference on Mobile Networks and Wireless Communications (ICMNWC)

摘要: OpenAI created and published ChatGPT, an AI chatbot, in November 2022. It was developed using supervised and reinforcement learning methods and is based on OpenAI’s GPT-3.5 and GPT -4 families of big language models. The creation of Artificial Intelligence has started a new era in the field of computer science. Although it was first established in 1950, by, now known as the father of artificial intelligence, John McCarthy, and before that by Alan Turning who explored its mathematical possibilities, Technology is generally ranked using a sigmoid curve, wherein you have a slow take off as the technology is invented, followed by an explosion of growth depending on how useful it is. That results in improvement in the technology as people find more real world practical applications for it, and then it reaches a stagnation point where you can’t improvise the tech anymore. We believe that Artificial Intelligence would be the start of a new sigmoid curve. Artificial Intelligence has now reached a new peak after the maturation of ChatGPT. Currently, AI has a gradual growth rate of 21 % per annum, creating a new market in this highly competitive industry. This paper represented the ChatGPT features, paradigm and data analysis process.

标题: Enhancing Autism Education: Exploring Interactive Videos and AI Integration for Effective Teaching

作者: Fatima Al Raeesi, Mariam K. Al Kuwaiti, Jose Berengueres

PubTime: 2023-11

Downlink: https://ieeexplore.ieee.org/document/10366485/

Journal: 2023 15th International Conference on Innovations in Information Technology (IIT)

中文摘要: 本研究项目建议使用互动视频来加强教师在自闭症儿童教育中的培训。针对自闭症学生的有效教学方法要求教师理解病情并采用量身定制的教学策略，包括调整作业，帮助有语言困难的人，以及利用视觉辅助工具更好地组织和集中注意力。传统的教师培训方法既昂贵又耗时。相比之下，交互式视频提供了一种主动、灵活的方式来访问培训内容，使教师能够动态地参与材料并控制他们的学习体验。未来的工作将探索人工智能驱动的ChatGPT的集成，以提供个性化的支持并创建一个动态的培训计划，目标是提高自闭症环境中的包容性和教育质量，同时使教师、学生和整个教育系统受益。

摘要: This research project proposes the use of interactive videos to enhance teachers’ training in educating autistic children. Effective teaching methods for students with autism require teachers to comprehend the condition and employ tailored instructional strategies, including adapting assignments, aiding those with language difficulties, and utilizing visual aids for better organization and focus. Traditional teacher training methods can be both expensive and time-consuming. In contrast, interactive videos provide a proactive and flexible way to access training content, empowering teachers to engage with the material dynamically and take control of their learning experiences. Future work will explore the integration of AI-driven ChatGPT to offer personalized support and create a dynamic training program, with the goal of improving inclusivity and educational quality in autism settings while benefiting teachers, students, and the education system at large.

标题: Elevating Employment Practices in Agricultural Corporations with Large Language Models and AI

作者: Samia A. Abu-Shanab, Ala Mughaid, Shadi AlZu'bi

PubTime: 2023-11

Downlink: https://ieeexplore.ieee.org/document/10375423/

Journal: 2023 Tenth International Conference on Social Networks Analysis, Management and Security (SNAMS)

中文摘要: 这项研究提案旨在利用现代人工智能技术，特别是ChatGPT，来转变农业信贷公司的人力资源。重点是人工智能在简化招聘、个性化员工体验、减轻偏见和增强决策方面的作用。通过利用人工智能和ChatGPT，这项研究旨在提高竞争力、效率和人力资源有效性。这些见解将指导农业领域人工智能驱动的人力资源解决方案，使组织和行业受益。在基础模型和大型语言模型（LLMs）时代，了解这些模型的潜力和局限性对这项研究至关重要，但由于资源限制，访问仅限于大型科技公司。全面研究和多学科合作对于揭示这些模型的能力和挑战至关重要。

摘要: This research proposal aims to leverage modern AI techniques, particularly ChatGPT, for transforming human resources in the Agriculture Credit Corporation. The focus is on AI’s role in streamlining recruitment, personalizing employee experiences, mitigating biases, and enhancing decision-making. By utilizing AI and ChatGPT, this study seeks to boost competitiveness, efficiency, and HR effectiveness. The insights will guide AI-driven HR solutions in agriculture, benefiting the organization and the industry. In the era of foundation models and large language models (LLMs), understanding these models’ potential and limitations is vital for this research, but access is limited to big tech companies due to resource constraints. Comprehensive research and multidisciplinary collaboration are essential to unravel the capabilities and challenges posed by these models.

标题: The Future of OpenAI Tools: Opportunities and Challenges for Human-AI Collaboration

作者: Kunal, Muskaan Rana, Jatin Bansal

PubTime: 2023-11

Downlink: https://ieeexplore.ieee.org/document/10424990/

Journal: 2023 2nd International Conference on Futuristic Technologies (INCOFT)

摘要: Artificial intelligence (AI) has the potential to transform numerous aspects of our lives, from healthcare and education to business and entertainment. OpenAI tools, such as ChatGPT and DALL-E 2, are at the forefront of this revolution, offering exciting opportunities for human-AI collaboration. However, there are also significant challenges to be overcome. This research paper explores the opportunities and challenges of using OpenAI tools for human-AI collaboration, with a focus on their potential applications, challenges, and key factors for successful collaboration. The paper presents case studies of successful collaborations, including their challenges and lessons learned, as well as future prospects of OpenAI tools. The research concludes with recommendations for continued research in human-AI collaboration, emphasizing the importance of responsible development and use of AI tools for the benefit of society.

== CLIP@ViT @ VLM @ visual model ==

标题: Optimized Design and Analysis of Clip Bonding Packaging for Avalanche Bipolar Junction Transistor

作者: Mosai Xu, Lin Liang, Xiaoxue Yan

PubTime: 2023-11

Downlink: https://ieeexplore.ieee.org/document/10395762/

Journal: 2023 IEEE 2nd International Power Electronics and Application Symposium (PEAS)

中文摘要: 为了满足雪崩双极结型晶体管（ABJT）对高开关速度日益增长的需求，并解决常用SOT23封装的散热问题，提出了一种采用夹键合的ABJT优化封装设计。本文基于有限元模拟进行了一系列分析，以研究新型夹式粘合封装的热机械性能。夹焊和引线焊封装之间的寄生参数、热和机械性能的比较表明，更大的焊接面积导致更低的电感和热阻，以及更好的散热性能和可靠性。还优化了铜片的厚度，并使用Hallberg-Peck模型进行了可靠性分析，证明了夹子键合封装的可靠性优于使用引线键合的传统SOT23封装。

摘要: To meet the increasing demand for high switch speed of avalanche bipolar junction transistors (ABJT) and to solve the heat dissipation problem of the commonly used SOT23 packaging, an optimized packaging design for ABJT using the clip bonding is proposed. In this paper, a series of analysis is conducted based on finite element simulation to study the thermomechanical performance of the novel clip bonding packaging. A comparison of the parasitic parameters, thermal and mechanical properties between clip bonding and wire bonding packaging demonstrates that the bigger bonding area leads to lower inductances and thermal resistance, as well as better heat dissipation performance and reliability. The thickness of the copper sheet is also optimized and the reliability analysis is conducted using the Hallberg-Peck model, demonstrating that the reliability of the clip bonding packaging is superior to the conventional SOT23 packaging using wire bonding.

标题: Video Question Answering Using Clip-Guided Visual-Text Attention

作者: Shuhong Ye, Weikai Kong, Chenglin Yao

PubTime: 2023-10

Downlink: https://ieeexplore.ieee.org/document/10222286/

Journal: 2023 IEEE International Conference on Image Processing (ICIP)

中文摘要: 视频和文本的跨模态学习在视频问答（VideoQA）中起着关键作用。在本文中，我们提出了一种视觉——文本注意机制，利用在大量通用领域语言——图像对上训练的对比语言——图像预训练（CLIP）来指导VideoQA的跨模态学习。具体来说，我们首先使用TimeSformer从目标应用领域提取视频特征，使用BERT从目标应用领域提取文本特征，并利用CLIP通过特定领域的学习从一般知识领域提取一对视觉文本特征。然后，我们提出了一种跨域学习方法，在目标域和一般域之间提取视觉和语言特征之间的注意信息。整合了一组剪辑引导的可视文本特征来预测答案。所提出的方法在MSVD-QA和MSRVTTQA数据集上进行了评估，并且优于最先进的方法。

摘要: Cross-modal learning of video and text plays a key role in Video Question Answering (VideoQA). In this paper, we propose a visual-text attention mechanism to utilize the Contrastive Language-Image Pre-training (CLIP) trained on lots of general domain language-image pairs to guide the cross-modal learning for VideoQA. Specifically, we first extract video features using a TimeSformer and text features using a BERT from the target application domain, and utilize CLIP to extract a pair of visual-text features from the general-knowledge domain through the domain-specific learning. We then propose a Cross-domain Learning to extract the attention information between visual and linguistic features across the target domain and general domain. The set of CLIP-guided visual-text features are integrated to predict the answer. The proposed method is evaluated on MSVD-QA and MSRVTTQA datasets and outperforms state-of-the-art methods.

标题: Eavesdropping Vulnerabilities in Optical Fiber Networks: Investigating Macro-Bending-Based Attacks Using Clip-on Couplers

作者: Vladimir Spurny, Petr Dejdar, Adrian Tomasov

PubTime: 2023-10

Downlink: https://ieeexplore.ieee.org/document/10328134/

Journal: 2023 International Workshop on Fiber Optics on Access Networks (FOAN)

中文摘要: 由于其复杂性，光缆长期以来被认为是防止窃听的安全介质。但是，有一些攻击方法允许窃听或拒绝服务。这些方法包括基于弯曲光纤的方法。危险在于能够在不中断路径的情况下实施攻击的可能性。为了便于这种攻击，可以使用夹式耦合器。这种耦合器被设计成允许不间断地窃听链路。在这项研究中，我们重点评估一种简单但商用的夹式耦合器的耦合效率。这项工作研究干扰到光链路的有效性，即复制干扰攻击，并通过模拟真实业务条件来测量它。为了更好地理解这些攻击的复杂性，我们进行了额外的测量来研究夹式耦合对极化态变化的影响。通过这项研究，我们强调了光纤越来越容易受到窃听企图，特别是那些使用基于宏的方法。通过评估商业上可用的夹式耦合的性能，我们为保护现代光通信网络的潜在风险和缓解策略提供了有价值的见解。

摘要: Fiber optic cables have long been considered a secure medium that prevents eavesdropping due to their complexity. However, there are attack methods that allow eavesdropping or denial of service. These include methods based on bending optical fibers. The danger resides in the possibility of being able to carry out an attack without having to interrupt the path. To facilitate such attacks, it is possible to use a clip-on coupler. This coupler is designed to allow eavesdropping on the link without interruption. In this study, we focus on evaluating the coupling efficiency of a simple but commercially available clip-on coupler. This work investigates the effectiveness of jamming into the optical link, i.e., replicating the jamming attack and measuring it by simulating real traffic conditions. To better understand the complexity of these attacks, we perform additional measurements to investigate the effect of clip-on coupling on changes in the polarization state. Through this research, we highlight the increasing vulnerability of optical fibers to eavesdropping attempts, especially those using macro-based methods. By evaluating the performance of commercially available clip-on coupling, we provide valuable insights into potential risks and mitigation strategies for securing modern optical communication networks.

标题: CLIP4VideoCap: Rethinking Clip for Video Captioning with Multiscale Temporal Fusion and Commonsense Knowledge

作者: Tanvir Mahmud, Feng Liang, Yaling Qing

PubTime: 2023-06

Downlink: https://ieeexplore.ieee.org/document/10097128/

Journal: ICASSP 2023 - 2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

中文摘要: 在本文中，我们基于大规模预训练的剪辑图像和文本编码器，结合多尺度时态推理和常识知识，提出了用于视频字幕的CLIP4VideoCap。除了对连续视频帧进行操作的剪辑图像编码器之外，我们还引入了一种基于知识提取的学习方案，旨在利用剪辑文本编码器从图像特征中生成丰富的文本知识。为了改进视频的时间推理，我们提出了一种多尺度时间融合方案，该方案从不同的时间窗口积累时间特征。此外，我们在字幕生成过程中集成了各种常识方面，通过在中间阶段从视频中提取常识特征，大大提高了字幕质量。结合这些策略，我们在基准MSR-VTT数据集上实现了最先进的性能，证实了我们的框架显著优于现有的方法。

摘要: In this paper, we propose CLIP4VideoCap for video captioning based on large-scale pre-trained CLIP image and text encoders together with multi-scale temporal reasoning and commonsense knowledge. In addition to the CLIP-image encoder operating on successive video frames, we introduce a knowledge distillation-based learning scheme that aims to exploit the CLIP-text encoder to generate rich textual knowledge from the image features. For improved temporal reasoning over the video, we propose a multi-scale temporal fusion scheme that accumulates temporal features from different temporal windows. In addition, we integrate various commonsense aspects in the caption generation which greatly enhances the caption quality by extracting the commonsense features from the video in the intermediate phase. Combining these strategies, we achieve state-of-the-art performance on the benchmark MSR-VTT dataset confirming that our framework significantly outperforms existing approaches.

标题: Can CLIP Share Image in Dialogue?

作者: Young-Jun Lee, Ho-Jin Choi

PubTime: 2023-02

Downlink: https://ieeexplore.ieee.org/document/10066624/

Journal: 2023 IEEE International Conference on Big Data and Smart Computing (BigComp)

中文摘要: 最近，许多研究构建了包含图像共享行为的多模态对话数据集，这对于在开放领域对话中增加与对话者的社会关系至关重要。在本文中，我们通过对（1）零镜头可转移性，（2）对话历史的影响，以及（3）鲁棒性进行各种实验，报告了CLIP可以理解对话历史和图像之间的对齐的实证结果。我们的实验表明，在多模态对话数据集上提高CLIP的零镜头性能是必要的。此外，剪辑模型受益于更多信息的文本（即对话历史），而不仅仅是最后的话语。

摘要: Recently, many studies have constructed multimodal dialogue datasets containing image-sharing behavior, which is vital to increase the social relationship with interlocutors in open-domain conversation. In this paper, we report the empirical results that CLIP can understand the alignment between the dialogue history and image by conducting various experiments for (1) zero-shot transferability, (2) the effect of dialogue history, and (3) robustness. Our experiments demonstrate that it is necessary for improving the zero-shot performance of CLIP on the multi-modal dialogue dataset. Additionally, the CLIP model is benefitted from more informative texts (i.e., dialogue history), not the last utterance only.

标题: ITMix: Image-Text Mix Augmentation for Transferring CLIP to Image Classification

作者: Tao Hong, Xiangyang Guo, Jinwen Ma

PubTime: 2022-10

Downlink: https://ieeexplore.ieee.org/document/9965292/

Journal: 2022 16th IEEE International Conference on Signal Processing (ICSP)

中文摘要: 像CLIP这样的跨模态模型的成功最近引发了研究人员对更好地理解不同模态之间的相互作用的兴趣。受CLIP有价值的零镜头图像分类实验的启发，本文重点研究了在传输CLIP时的数据扩充，以微调下游分类任务。像Mixup或CutMix这样的Mix系列是一种有效的数据增强方法，它通过在不同样本之间进行插值来生成新图像。不同于普通的mix系列只专注于图像模态的增强，我们打算同时混合图像和文本模态，命名为ITMix。通过这种方式，将创建更丰富的匹配图像——文本对。对于ITMix的实现，提出了有效的匹配损失微调和软一对多映射。实验结果验证了该方法在不同图像分类基准（CIFAR-10、CIFAR-100、Food-101等）上的优越性。

摘要: The success of cross-modal models like CLIP has sparked researchers’ interest in better understanding the interaction between different modalities recently. Inspired by the valuable zero-shot image classification experiment of CLIP, we focus on data augmentation when transferring CLIP for finetuning downstream classification tasks in this paper. Mix series like Mixup or CutMix is an effective data augmentation method that generates new images by interpolating between different samples. Different from the common mix series which only concentrates on augmentation of image modality, we intend to mix image and text modalities simultaneously, named ITMix. In this way, more abundant matched image-text pairs would be created. For the implementation of ITMix, effective fine-tuning with match loss and soft one-to-more mapping are proposed. The experimental results verify the outperformance of our proposed method in terms of accuracy on different image classification benchmarks: CIFAR-10, CIFAR-100, Food-101, etc.

== diffusion policy@diffusion formulation@diffusion model ==

标题: Diffusion Co-Policy for Synergistic Human-Robot Collaborative Tasks

作者: Eley Ng, Ziang Liu, Monroe Kennedy

PubTime: 2024-01

Downlink: https://ieeexplore.ieee.org/document/10310116/

Journal: IEEE Robotics and Automation Letters

Project: http://www.w3.org/1998/Math/MathML|http://www.w3.org/1999/xlink|http://www.w3.org/1998/Math/MathML|http://www.w3.org/1999/xlink|

中文摘要: 多模态人类行为建模一直是提高人类和机器人之间交互水平的关键障碍，特别是对于协作任务。我们的<bold xmlns：mml=”http：//www.w3.org/1998/Math/MathML”xmlns：xlink=”http：//www.w3.org/1999/xlink”>关键见解是，用于人——机器人协作任务的有效、学习的机器人策略必须能够表达高度的多模态，以时间一致的方式预测动作，并识别人类动作的大范围频率，以便在控制回路中与人类无缝集成。我们提出了扩散协同策略，这是一种在测试期间规划与人类协同良好的行动序列的方法。协同策略通过基于Transformer model的扩散模型预测联合人——机器人动作序列，该模型在协作人——人演示的数据集上训练，并在后退地平线控制框架中直接执行机器人动作。我们在模拟和真实环境中证明了该方法在人在回路中的人——机器人桌子搬运任务上优于其他最先进的学习方法。此外，我们定性地强调了令人信服的机器人行为，这些行为证明了真正的人——机器人协作的证据，包括相互适应、共享任务理解、领导转换和低水平的由异议引起的浪费互动力。

摘要: Modeling multimodal human behavior has been a key barrier to increasing the level of interaction between human and robot, particularly for collaborative tasks. Our
key insight
is that an effective, learned robot policy used for human-robot collaborative tasks must be able to express a high degree of multimodality, predict actions in a temporally consistent manner, and recognize a wide range of frequencies of human actions in order to seamlessly integrate with a human in the control loop. We present Diffusion Co-policy, a method for planning sequences of actions that synergize well with humans during test time. The co-policy predicts joint human-robot action sequences via a Transformer-based diffusion model, which is trained on a dataset of collaborative human-human demonstrations, and directly executes the robot actions in a receding horizon control framework. We demonstrate in both simulation and real environments that the method outperforms other state-of-art learning methods on the task of human-robot table-carrying with a human in the loop. Moreover, we qualitatively highlight compelling robot behaviors that demonstrate evidence of true human-robot collaboration, including mutual adaptation, shared task understanding, leadership switching, and low levels of wasteful interaction forces arising from dissent.

标题: Combined Diffusion Adaptation on Adaptive Leaky Criterion and Orthogonal Gradient Algorithm

作者: Suchada Sitjongsataporn, Piyaporn Nurarak

PubTime: 2023-05

Downlink: https://ieeexplore.ieee.org/document/10153328/

Journal: 2023 20th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology (ECTI-CON)

中文摘要: 本文提出了一种基于正交梯度的算法的组合扩散策略，该算法使用自适应平均泄漏准则。混合节点准则由其自身抽头权重向量的正交投影矩阵描述，而其他连接节点的信息不受干扰。采用自适应泄漏算法实现了快速收敛和低复杂度。通过最小化混合节点成本函数，根据分布式网络上的先适应后组合策略来验证分布式估计。系统辨识的统计实验结果表明，所提出的算法可以以均方误差准则的形式提供有希望的结果。

摘要: This paper presents a combined diffusion policy on the orthogonal gradient-based algorithm using adaptive averaging leaky criterion. A mixed-node criterion is described by orthogonal projection matrix of its own tap-weight vector, while information of other connected nodes are undisturbed. An adaptive leaky algorithm is applied for fast convergence with low complexity. By minimising a mixed-node cost function, the distributed estimation is verified in terms of adapt-then-combine strategy over the distributed network. Statistical experimental results on system identification examine that a proposed algorithm can provide promising results in form of mean square error criterion.

标题: CO2 Reduction Potential by Putting Electric Vehicles into Operation in Phu Quoc Island, Viet Nam

作者: Hoang-Phuong Nguyen, Viet-Cuong Vo, Tan-Dong Le

PubTime: 2019-07

Downlink: https://ieeexplore.ieee.org/document/8823377/

Journal: 2019 International Conference on System Science and Engineering (ICSSE)

中文摘要: 本文重点研究CO2减排的潜力，通过将电动汽车投入运行来取代传统的燃料车辆，并为实现越南建江富国的可持续发展目标做出贡献。该论文采用了富国2020-2030年期间与人口增长、人均收入预测、化石燃料车辆增长计算相关的大量规划数据。此外，本文还提出了一个电动汽车市场的技术扩散模型和不同的方法，参考了世界各国促进电动汽车发展的政策框架。在此基础上，作者提出了许多电动汽车方案，以渗透市场并确定CO2到2030年减排的潜力。同时，作者发布了政策框架和建议，以便研究成果能够在实践中尽快实施。预计到2030年，化石燃料汽车的排放量约为79,901吨。通过政府在发展电动汽车方面的政策框架建议，预测CO2排放减少量将从13,859吨减少到16,489吨C02/年，相当于在各种情况下减少17-20%。研究结果显示，如果拟议的政策框架尽快实施，并在富国引入电动汽车以取代传统的燃料汽车是必要和可行的，不仅在减少CO2排放的能力方面，而且有助于当地发展成为区域和国际绿色旅游目的地。

摘要: This paper focuses on studying the potential of CO
₂
emission reduction by putting electric vehicles into operation to replace traditional fuel-use vehicles and to contribute to achieving sustainable development goals in Phu Quoc, Kien Giang, Vietnam. The paper employed numerous planning data of Phu Quoc for the period of 2020 - 2030 related to the population increase, forecasted income per capita, computed growth of vehicles by fossil fuels. Also, the paper has launched a technology diffusion model and different approaches on electric vehicle market, refer to policy frameworks of many countries around the world to promote electric vehicles development. On that basis, the author proposes numerous electric vehicle scenarios to penetrate the market and determine the potential of CO
₂
emission reduction by 2030. At the same time, the author releases suggestions on policy frameworks and recommendations so that the research results could be soon implemented in practice. With the forecast until 2030, the emission resulted by fossil fuel vehicles is about 79,901 tons. Through the policy framework proposals for the government in the development of electric vehicles, it is forecasted that CO
₂
emission reductions will be from 13,859 to 16,489 tons of C0
₂
/year, corresponding to a reduction of 17-20% for scenarios. Research results show that if the proposed policy frameworks are implemented soon and the introduction of electric vehicles to replace traditional fuel vehicles in Phu Quoc is necessary and feasible not only in terms of ability reduce CO
₂
emissions, but also contribute to local development to become a regional and international green tourism destination.

标题: Exploring the Influencing Factors of the Diffusion of Electric Vehicles Sharing Service in Shanghai Using Fuzzy Cognitive Maps

作者: Peng-yun TIAN, Tie-ju MA, Hong-bin YAN

PubTime: 2018-08

Downlink: https://ieeexplore.ieee.org/document/8745202/

Journal: 2018 International Conference on Management Science and Engineering (ICMSE)

中文摘要: 随着环境污染、交通堵塞和停车问题的加剧，人们开始寻找低碳出行方式。同时，移动互联网、第三方支付、GPS定位等信息技术的发展，为出行共享服务的出现提供了基础。旅游共享服务作为一个新兴产业，在城市中的推广是企业和政府关注的焦点。本文以电动汽车共享服务为例进行研究。从政府、企业和用户三个方面出发，收集并分析了影响电动汽车共享服务扩散的因素。模糊认知图是一种分析复杂决策问题的技术，我们在本研究中使用它来识别影响电动汽车共享服务扩散的关键因素。扩散的模拟是基于上海电动汽车共享服务的现状。结果表明，不同的利益相关者群体有不同的观点，并确定不同的因素作为电动汽车共享服务扩散的驱动因素。模拟表明，基于模糊认知图开发的模型对政治因素更敏感。

摘要: With the increasing of environmental pollution, traffic jams and parking problems, people have begun to look for low-carbon travel ways. At the same time, the development of information technologies such as mobile internet, third-party payment, and GPS positioning, has provided a foundation for the emergence of travel sharing services. As a new industry, the diffusion of travel sharing services in cities is the focus of enterprises and government. This article takes the electric vehicle sharing service as an example to carry out the study. Starting from government, enterprises and the users, we collect and analyze the factors that affect the diffusion of the electric vehicle sharing service. Fuzzy cognitive map is a technique for analyzing complex decision-making problems and we use it to identify the key factors affecting the diffusion of electric vehicle sharing service in this study. Simulations of the diffusion is based on the current state of electric vehicle sharing service in Shanghai. The results show that different stakeholder groups have different views and identify different factors as driving factors for the diffusion of electric vehicle sharing services. The simulations have indicated that the model developed based on fuzzy cognitive maps is more sensitive to political factors.

标题: Regulatory Policy Awareness and Environmental Supply Chain Cooperation in China: A Regulatory-Exchange-Theoretic Perspective

作者: Qinghua Zhu, Joseph Sarkis, Kee-Hung Lai

PubTime: 2018-02

Downlink: https://ieeexplore.ieee.org/document/8013055/

Journal: IEEE Transactions on Engineering Management

中文摘要: 一些发达国家颁布了生产者责任延伸条例，促进环境供应链合作（ESCC）做法在制造商中的传播。中国等发展中国家也采取了类似但总体上灵活和自愿的法规和政策。使用以监管方面为重点的交换理论作为理论视角，本文提出了一些命题，以检验制造商对自愿环境监管政策的认识是否不同，以及与ESCC实践采用的关系。对308份答复的聚类分析和多元方差分析结果确定了三类中国制造商对环境监管政策的认识。这三类包括精明、专注和无知的制造商。研究发现，具有较高环境监管意识的制造商倾向于更深入地实施ESCC实践。分层回归分析进一步用于检验监管政策意识和ESCC实践之间的关系。只有精明的制造商了解国内监管政策才会对绿色采购产生积极影响。回归结果显示，国内监管政策意识与消费者对环境问题的合作之间存在非线性关系，从不明智的制造商略微正向，专注的制造商略微负向，到精明的制造商显著正向。

摘要: Some developed countries have enacted extended producer responsibility regulations that facilitate the diffusion of environmental supply chain cooperation (ESCC) practices among manufacturers. Developing countries, such as China, have adopted similar but generally flexible and voluntary regulations and policies. Using exchange theory with a focus on regulatory aspects as the theoretical lens, this paper develops propositions to examine if awareness of voluntary environmental regulatory policies is different among manufacturers as well as the relationship to ESCC practices adoption. Results from cluster analysis and multivariate analysis of variance for 308 responses identify three categories of Chinese manufacturers with respect to their awareness of environmental regulatory policies. These three categories include savvy, attentive, and nescient manufacturers. It was found that manufacturers characterized with higher environmental regulatory awareness tend to implement ESCC practices more intensively. Hierarchical regression analysis was further used to examine the relationship between awareness of regulatory policies and ESCC practices. Awareness of domestic regulatory policies has positive effects on green purchasing only for savvy manufacturers. Regression results show a nonlinear relationship between awareness of domestic regulatory policies and customer cooperation with environmental concerns, from slightly positive for nescient manufacturers and slightly negative for attentive manufacturers to significantly positive for savvy manufacturers.

标题: Sustainable electric vehicle - prosumer framework and policy mix

作者: Kirsi Kotilainen, Saku J. Mäkinen, Jussi Valta

PubTime: 2017-12

Downlink: https://ieeexplore.ieee.org/document/8378406/

Journal: 2017 IEEE Innovative Smart Grid Technologies - Asia (ISGT-Asia)

摘要: Electric vehicles have pro-environmental advantages compared to traditional automobiles, or even hybrids: they can help reducing pollution and noise levels locally, and greenhouse gas emissions globally. However, there are still many challenges that the electric vehicles must overcome before reaching level of diffusion that can have significant impact on sustainability. This paper evaluates combined sustainability of electric vehicle and small-scale energy production. We propose a framework for sustainable electric vehicle - energy prosumer integration and outline a policy mix that is needed to support adoption of both renewable energy technologies and electric vehicles.

== Visual Navigation@VLN @ Visual Language Navigation ==

标题: Gateway Crossing Method for Unmanned Aerial Vehicles Based on Lightweight Corner-Assisted Localization and Visual-Inertial Navigation

作者: Xudong Liu, Yufan Peng, Huangchao Yu

PubTime: 2023-07

Downlink: https://ieeexplore.ieee.org/document/10339505/

Journal: 2023 2nd International Conference on Machine Learning, Cloud Computing and Intelligent Mining (MLCCIM)

中文摘要: 本文提出了一种基于传统视觉惯性里程表的微型无人机轻量级网关穿越方法，该方法集成了某些地图先验信息。首先介绍了一种融合入口拐角信息的视觉惯性导航系统。该系统包括角点辅助定位方法、误识别检测机制和融合方法。进行了网关穿越仿真和飞行试验。结果表明，所提出的无人机网关穿越方法在准确性、速度和轻量级方面具有优势。与VINS-MONO相比，该定位系统的平均轨迹误差分别降低了20.1%（模拟）和34.3%（物理测试）。所提出的方法在内存和CPU利用率节省方面优于基于EGO-Planner的网关交叉方法，内存使用量减少29.0%（模拟）和7.7%（物理测试），CPU使用量减少21.9%（模拟）和17.7%（物理测试）。

摘要: This paper proposes a lightweight gateway crossing method for micro-UAVs that integrates certain map prior information, based on a traditional visual-inertial odometer. A visual-inertial navigation system which fuses the gateway corner information is firstly introduced. This system includes a corner-assisted localization method, misrecognition detection mechanism and fusion method. The gateway crossing simulation and flight tests are carried out. The results show advantages in the accuracy, speed and lightweight of the proposed gateway crossing method for UAVs. The average trajectory error of the positioning system decreases by 20.1 % (simulation) and 34.3% (physical test) compared to those of VINS- Mono. The proposed method outperforms the gateway crossing method based on EGO-Planner in terms of memory and CPU utilization savings, with 29.0% (simulation) and 7.7% (physical test) memory usage reduction, and 21.9% (simulation) and 17.7% (physical test) CPU usage reduction.

标题: LiDAR/Visual SLAM-Aided Vehicular Inertial Navigation System for GNSS-Denied Environments

作者: Nader Abdelaziz, Ahmed El-Rabbany

PubTime: 2022-12

Downlink: https://ieeexplore.ieee.org/document/10019210/

Journal: 2022 5th International Conference on Communications, Signal Processing, and their Applications (ICCSPA)

中文摘要: GNSS挑战环境中的大多数导航系统依赖于GNSS/INS组合导航系统，INS有可能在GNSS短暂中断期间提供可靠的定位。然而，在全球导航卫星系统信号长时间中断的情况下，系统的性能将完全取决于INS解决方案，这可能导致随时间的显著漂移。因此，增加更多的机载传感器对于减轻GNSS/INS系统的限制，从而提高导航系统的鲁棒性至关重要。本研究提出了一种使用扩展卡尔曼滤波器（EKF）的INS、激光雷达同步定位和绘图（SLAM）和可视SLAM之间的松耦合（LC）集成。开发的集成导航系统在原始KITTI数据集的住宅和公路驾驶部分进行了测试，该数据集在特征密度和驾驶速度方面模拟了各种驾驶户外环境。在这两种情况下，都强制执行完全人为的GNSS中断。结果表明，所提出的INS/LiDAR/visual SLAM集成系统的性能明显优于仅使用INS的系统。所提出的组合导航系统在东、北和向上方向分别产生了大约95%、87%和53%的均方根误差（RMSE）的平均降低。最后，所提出的算法优于所考虑的最先进的激光雷达SLAM算法。

摘要: Most navigation systems in GNSS-challenged environments rely on GNSS/INS integrated navigation system, with the INS potentially providing reliable positioning during short GNSS outages. However, in the event of a prolonged GNSS signal outage, the performance of the system will be solely dependent on the INS solution, which can lead to significant drift over time. As a result, adding more onboard sensors is crucial to mitigate the limitation the GNSS/INS systems, and thereby increase the robustness of the navigation system. This study proposes a loosely-coupled (LC) integration between the INS, LiDAR simultaneous localization and mapping (SLAM), and visual SLAM using an extended Kalman filter (EKF). The developed integrated navigation system is tested on the residential and highway drive segments of the raw KITTI dataset, which simulates various driving outdoor environments in terms of feature density and driving speed. In both cases, a complete artificial GNSS outage is enforced. The results show that the proposed INS/LiDAR/visual SLAM integrated system drastically outperforms the use of INS only. The proposed integrated navigation system yielded an average reduction in the root-mean-square error (RMSE) of approximately 95%, 87%, and 53%, in the east, north, and up directions, respectively. Finally, the proposed algorithm outperformed considered state-of-the-art LiDAR SLAM algorithms.

标题: Decoupled Right Invariant Error States for Consistent Visual-Inertial Navigation

作者: Yulin Yang, Chuchu Chen, Woosik Lee

PubTime: 2022-04

Downlink: https://ieeexplore.ieee.org/document/9669116/

Journal: IEEE Robotics and Automation Letters

中文摘要: 不变扩展卡尔曼滤波器（IEKF）被证明保持了视觉惯性导航系统（VINS）的可观测性，并且适合于一致的估计器设计。然而，如果特征保持在状态向量中，IEKF的传播将变得更加计算昂贵，因为这些特征涉及协方差传播。为了解决这个问题，我们提出了两种新的算法，通过利用不变状态表示来保持系统的一致性，并通过将特征从协方差传播中解耦来确保效率。第一种算法将右不变误差状态与第一估计雅可比（FEJ）技术相结合，通过将特征从李群表示中解耦，并利用FEJ进行一致估计。第二种算法是专门为基于滑动窗口滤波器的vin设计的，因为它将特征与活动克隆姿态相关联，而不是当前IMU状态，用于李群表示。还提出了一种新的伪锚点改变算法，以保持状态向量中的特征比窗口跨度长。实现了基于解耦的左右不变误差的VINS方法，以进行完整的比较。在三个模拟轨迹上进行了大量的蒙特卡罗模拟，并在TUM-VI数据集上进行了真实世界的评估，以验证我们的分析，并证明所提出的算法可以实现比使用FEJ的基于滤波器的VINS算法更高的精度。

摘要: The invariant extended Kalman filter (IEKF) is proven to preserve the observability property of visual-inertial navigation systems (VINS) and suitable for consistent estimator design. However, if features are maintained in the state vector, the propagation of IEKF will become more computationally expensive because these features are involved in the covariance propagation. To address this issue, we propose two novel algorithms which preserve the system consistency by leveraging the invariant state representation and ensure efficiency by decoupling features from covariance propagation. The first algorithm combines right invariant error states with first-estimates Jacobian (FEJ) technique, by decoupling the features from the Lie group representation and utilizing FEJ for consistent estimation. The second algorithm is designed specifically for sliding-window filter-based VINS as it associates the features to an active cloned pose, instead of the current IMU state, for Lie group representation. A new pseudo-anchor change algorithm is also proposed to maintain the features in the state vector longer than the window span. Both decoupled right- and left-invariant error based VINS methods are implemented for a complete comparison. Extensive Monte-Carlo simulations on three simulated trajectories and real world evaluations on the TUM-VI datasets are provided to verify our analysis and demonstrate that the proposed algorithms can achieve improved accuracy than a state-of-art filter-based VINS algorithm using FEJ.

标题: Loose Coupled Initialization Method for Visual Inertial Navigation in Outdoor Low Altitude Flight Environment

作者: Yibin Wang, Chengwei Yang, Cheng Zhang

PubTime: 2021-11

Downlink: https://ieeexplore.ieee.org/document/9657780/

Journal: 2021 6th International Conference on Robotics and Automation Engineering (ICRAE)

中文摘要: 在户外低空飞行环境中，惯性导航自对准初始化方法很难应用于动态环境，因为非自主对准方法在大失准角的情况下不准确，并且使用视觉信息辅助惯性导航系统初始化的方法依赖于视界提取算法或人工标志的提前部署，不够灵活。针对上述问题，提出了一种基于视觉信息的松耦合惯性导航系统初始化方法。首先，在视觉前端将特征点分割成天空端和地面端。并且使用来自运动的视觉结构来计算相机坐标系中的相机姿态。然后，将摄像机姿态与加速度和角速度测量值的重新积分对齐，以获得初始化姿态。通过半物理仿真系统在三种情况下进行低空飞行仿真，验证了算法的性能。

摘要: In the outdoor low altitude flight environment, the inertial navigation self-alignment initialization method is difficult to apply to the dynamic environment, since the non-autonomous alignment method is not accurate in the case with large misalignment angles, and the method of using visual-information-assisted inertial navigation system initialization relies on the horizon extraction algorithm or the advance deployment of artificial landmarks, which is not flexible enough. To solve the problems above, a loose coupled inertial navigation system initialization method is proposed aided by visual information. First, the feature points are segmented into sky end and ground end at the vision front end. And the camera poses in the camera coordinate system are calculated using the visual structure from motion. Then the camera poses and the re-integration of acceleration and angular velocity measurements are aligned to obtain the initialized poses. The performance of our algorithm is verified by performing a low altitude flight simulation in three cases with a semi-physical simulation system.

标题: Suppression of Noise in Visual Navigation Systems

作者: Maksim G. Lutsky, Viktor M. Sineglazov, Vitaly S. Ishchenko

PubTime: 2021-10

Downlink: https://ieeexplore.ieee.org/document/9615405/

Journal: 2021 IEEE 6th International Conference on Actual Problems of Unmanned Aerial Vehicles Development (APUAVD)

中文摘要: 在这项工作中，考虑了大气噪声的影响。这种天气对图像的影响大大降低了视觉导航系统的精度，视觉导航系统识别地标以确定无人机的当前坐标。为了提高视觉导航系统的精度，对图像中的噪声进行滤除是非常重要的。提出了利用卷积神经网络对图像中的恶劣天气条件进行过滤。文中还对图像中不同噪声的数学模型进行了评述。视觉导航系统在没有全球定位系统信号的区域提供无人机的自主飞行。因此，在噪声和恶劣天气条件下，提供最佳质量的图像滤波对于视觉导航系统的精度和性能非常重要。现代计算机视觉算法无法完全去除图像上的噪声。在这种情况下，卷积神经网络具有显著的优势和方法被考虑。

摘要: In this work atmosphere noise influence is considered. Such weather effects on the images greatly decrease accuracy of visual navigation systems, which recognize landmarks to determine current coordinates of unmanned aerial vehicle. It is very important to filter noise on images to increase accuracy of visual navigation system. To filter bad weather conditions on the images using convolution neural networks is proposed. Also math models of different noises on the images is reviewed. Visual navigation system provides autonomous flight of unmanned aerial vehicle in areas where are Global Positioning System signals is absent. For this reason providing best quality of the images due to noise and bad weather conditions filtering is very important for visual navigation system accuracy and performance. Modern computer vision algorithms can't remove full noise on the image. In this case convolution neural networks have a significant advantage and approach is considered.

标题: Multi-Intelligent Vehicle Cooperative Formation Control Method based on Visual Navigation

作者: Jianglin Lu

PubTime: 2020-12

Downlink: https://ieeexplore.ieee.org/document/9332843/

Journal: 2020 IEEE International Conference on Industrial Application of Artificial Intelligence (IAAI)

中文摘要: 传统的多智能体车辆协同编队控制方法中存在大量路径信息噪声，导致动态稳定性差。利用视觉导航系统中的图像采集卡获取多辆智能车辆的路径信息，提高图像质量，提取路径边缘，获得完整的路径特征，计算目标智能车辆的预期速度和预期位置，并在改进的原始控制协议上执行，实现多辆智能车辆的协同编队控制。实验结果表明，所设计的基于视觉导航的多智能体车辆协同编队控制方法对方向和加速度响应及时有效，速度一致性保持在较高水平，表明控制方法的动态稳态性能得到了改善。

摘要: There are many path information noises in the traditional multi-agent vehicle cooperative formation control method, which leads to poor dynamic stability. Use the image capture card in the visual navigation system to obtain the path information of multiple intelligent vehicles, enhance the image quality, extract the edge of the path, obtain the complete path feature, calculate the expected speed and expected position of the target intelligent car, and perform it on the original control protocol Improved to realize the cooperative formation control of multiple intelligent vehicles. The experimental results show that the designed multi-agent vehicle cooperative formation control method based on visual navigation has a timely and effective response to direction and acceleration, and the speed consistency is maintained at a high level, which shows that the dynamic steady-state performance of the control method has been improved.

专属领域论文订阅

VX关注{晓理紫}，每日更新论文，如感兴趣，请转发给有需要的同学，谢谢支持

如果你感觉对你有所帮助，请关注我，每日准时为你推送最新论文。

在这里插入图片描述