检索结果-内蒙古大学图书馆

Sentence Representation Learning with Generative Objective rather than Contrastive Objective

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Wu, Bohong Zhao, Hai Department of Computer Science and Engineering Shanghai Jiao Tong University China Key Laboratory of Shanghai Education Commission for Intelligent Interaction and Cognitive Engineering Shanghai Jiao Tong University China

Though offering amazing contextualized token-level representations, current pre-trained language models take less attention on accurately acquiring sentence-level representation during their self-supervised pre-training. However, contrastive objectives which dominate the current sentence representation learning bring little linguistic interpretability and no performance guarantee on downstream semantic tasks. We instead propose a novel generative self-supervised learning objective based on phrase reconstruction. To overcome the drawbacks of previous generative methods, we carefully model intra-sentence structure by breaking down one sentence into pieces of important phrases. Empirical studies show that our generative learning achieves powerful enough performance improvement and outperforms the current state-of-the-art contrastive methods not only on the STS benchmarks, but also on downstream semantic retrieval and reranking tasks. Our code is available at https://***/chengzhipanpan/PaSeR. Copyright © 2022, The Authors. All rights reserved.

关键词： Benchmarking

Research on Autonomous Robots Navigation based on Reinforcement Learning

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Wang, Zixiang Yan, Hao Wang, Yining Xu, Zhengjia Wang, Zhuoyue Wu, Zhizhong College of Engineering and Computer Science Syracuse University SyracuseNY United States Human Factors in Information Design Bentley University WalthamMA United States SATM Cranfield University Cranfield United Kingdom Department of Electrical Engineering and Computer Sciences University of California Berkeley BerkeleyCA United States Mountain View CA United States

Reinforcement learning continuously optimizes decision-making based on real-time feedback reward signals through continuous interaction with the environment, demonstrating strong adaptive and self-learning capabilities. In recent years, it has become one of the key methods to achieve autonomous navigation of robots. In this work, an autonomous robot navigation method based on reinforcement learning is introduced. We use the Deep Q Network (DQN) and Proximal Policy Optimization (PPO) models to optimize the path planning and decision-making process through the continuous interaction between the robot and the environment, and the reward signals with real-time feedback. By combining the Q-value function with the deep neural network, deep Q network can handle high-dimensional state space, so as to realize path planning in complex environments. Proximal policy optimization is a strategy gradient-based method, which enables robots to explore and utilize environmental information more efficiently by optimizing policy functions. These methods not only improve the robot's navigation ability in the unknown environment, but also enhance its adaptive and self-learning capabilities. Through multiple training and simulation experiments, we have verified the effectiveness and robustness of these models in various complex scenarios. Copyright © 2024, The Authors. All rights reserved.

关键词： Adversarial machine learning

Automatic detection and classification algorithm of thyroid nodules in CT images based on YOLOv5s

学校读者我要写书评

暂无评论

Automatic detection and classification algorithm of thyroid ...

International Conference on computer and Communications (ICCC)

作者： Ruifeng Guo Lei Xu Jiahui Guo Hongliang Wang Shenyang Institute of Computing Technology Chinese Academy of Sciences University of Chinese Academy of Sciences Liaoning Province Human-Computer Interaction System Engineering Research Center Based on Digital Twin Shenyang China Department of Endocrinology and Metabolism The First Hospital of China Medical University Shenyang China

Thyroid nodular lesions are one of the most common diseases in the population. As the number of patients increases, it is clinically important to provide physicians with an accurate early aid to determine the benignity and malignancy of thyroid nodules in order to avoid unnecessary surgery and reduce patient burden. In this paper, we propose an improved Yolov5s-based target detection algorithm for benign and malignant thyroid nodules for thyroid nodule classification and identification. Replacing the backbone network of Yolov5s with ResNet18 to reduce the complexity of the model and the number of parameters; introducing the CBAM attention mechanism into the Yolov5s algorithm to improve the ability to extract features; Modify the spatial pyramid pooling module SPP to SimSPPF to improve the training speed of the algorithm; replace GIoU Loss with CIoU-Focal Loss as the loss function of the algorithm to improve the localization accuracy while increasing the regression rate of the bounding box. The experimental results show that the improved Yolov5s achieved an average accuracy of 92.9%, which is 2.9 percentage points higher than the average accuracy of the original algorithm, and can effectively improve the accuracy and efficiency of nodule detection.

关键词：

SKU-Patch: Towards Efficient Instance Segmentation for Unseen Objects in Auto-Store

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Yang, Biqi Tang, Weiliang Gao, Xiaojie Li, Xianzhi Liu, Yun-Hui Fu, Chi-Wing Heng, Pheng-Ann The Department of Computer Science and Engineering The Chinese University of Hong Kong Hong Kong Guangdong-Hong Kong-Macao Joint Laboratory of Human-Machine Intelligence-Synergy Systems Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences China The School of Computer Science and Technology Huazhong University of Science and Technology China The Department of Mechanical and Automation Engineering The Chinese University of Hong Kong Hong Kong

In large-scale storehouses, precise instance masks are crucial for robotic bin picking but are challenging to obtain. Existing instance segmentation methods typically rely on a tedious process of scene collection, mask annotation, and network fine-tuning for every single Stock Keeping Unit (SKU). This paper presents SKU-Patch, a new patch-guided instance segmentation solution, leveraging only a few image patches for each incoming new SKU to predict accurate and robust masks, without tedious manual effort and model re-training. Technical-wise, we design a novel transformer-based network with (i) a patch-image correlation encoder to capture multi-level image features calibrated by patch information and (ii) a patch-aware transformer decoder with parallel task heads to generate instance masks. Extensive experiments on four storehouse benchmarks manifest that SKU-Patch is able to achieve the best performance over the state-of-the-art methods. Also, SKU-Patch yields an average of nearly 100% grasping success rate on more than 50 unseen SKUs in a robot-aided auto-store logistic pipeline, showing its effectiveness and practicality. Copyright © 2023, The Authors. All rights reserved.

关键词： Benchmarking

Commissioning Development: Grantmaking, Community Voices, and their Implications for ICTD 2022

学校读者我要写书评

暂无评论

Commissioning Development: Grantmaking, Community Voices, an...

International Conference on Information & Communication Technologies and Development 2022

作者： Manika Saha Tom Bartindale Sharifa Sultana Gillian Oliver Dan Richardson Shakuntala Haraksingh Thilsted Syed Ishtiaque Ahmed Patrick Olivier Action Lab Human-centric computing Faculty of IT Monash University Australia Department of Information Science Cornell University USA Faculty of IT Monash University Australia Global Public Health Nutrition WorldFish Malaysia Department of Computer Science University of Toronto Canada

ISBN: (纸本)9781450397872

While information and communication technology for development (ICTD) researchers have prioritized advocating for community voices in innovation design and development, we have limited insights into how community voices are incorporated by the high-level decision-makers who fund and initiate development projects and programs in the Global South. Indeed, understanding local communities’ voices (expressions of needs, challenges, and priorities) in tailoring effective development projects for sustainable development is widely considered an unmet goal. Using a qualitative survey of eight decision-makers (including grantmaking donors, central governments and INGOs) we explored a number of key factors, including national and global political climates, insider-outsider interactions, and evidence-based approaches that influence the high-level decision making process, workflows, and perceptions of community voice in project commissioning within Bangladesh’s public health nutrition development arena. Our findings reveal the tensions that arise among high-level decision-makers, and highlight the challenges associated with connecting with communities during development project design and implementation. We suggest broader implications and design opportunities for inventive project commissioning approaches to bridge the gap between communities and decision-makers. Our findings are of potential value for ICTD and HCI4D researchers interested in sustainable innovation and understanding and participating in the complex workflows of the project commissioning process in sustainable global development.

关键词：

miR-204 ameliorates osteoarthritis pain by inhibiting SP1-LRP1 signaling and blocking neuro-cartilage interaction

学校读者我要写书评

暂无评论

Bioactive Materials 2023年第8期26卷 425-436页

作者： Ke Lu Qingyun Wang Liuzhi Hao Guizheng Wei Tingyu Wang William WLu Guozhi Xiao Liping Tong Xiaoli Zhao Di Chen Research Center for Computer-aided Drug Discovery Shenzhen Institute of Advanced TechnologyChinese Academy of SciencesShenzhen518055China Faculty of Pharmaceutical Sciences Shenzhen Institute of Advanced TechnologyShenzhen518055China Research Center for Human Tissues and Organs Degeneration Shenzhen Institute of Advanced TechnologyChinese Academy of SciencesShenzhen518055China Department of Pharmacy Shanghai Ninth People’s HospitalShanghai Jiao-Tong University School of MedicineShanghai200011China Department of Orthopaedics and Traumatology The University of Hong KongHong Kong 999077China School of Medicine Southern University of Science and TechnologyShenzhen518055China

Osteoarthritis(OA)is a painful degenerative joint disease and is the leading cause of chronic disability among elderly *** improve the quality of life for patients with OA,the primary goal for OA treatment is to relieve the *** OA progression,nerve ingrowth was observed in synovial tissue and articular *** abnormal neonatal nerves act as nociceptors to detect OA pain *** molecular mechanisms for transmitting OA pain in the joint tissues to the central nerve system(CNS)is currently *** miR-204 has been demonstrated to maintain the homeostasis of joint tissues and have chondro-protective effect on OA ***,the role of miR-204 in OA pain has not been *** this study,we investigated interactions between chondrocytes and neural cells and evaluated the effect and mechanism of miR-204 delivered by exosome in the treatment of OA pain in an experimental OA mouse *** findings demonstrated that miR-204 could protect OA pain by inhibition of SP1-LDL Receptor Related Protein 1(LRP1)signaling and blocking neuro-cartilage interaction in the *** studies defined novel molecular targets for the treatment of OA pain.

关键词： miR-204 LRP1 Neuro-cartilage interaction Osteoarthritis Pain

Learning Better Masking for Better Language Model Pre-training

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Yang, Dongjie Zhang, Zhuosheng Zhao, Hai Department of Computer Science and Engineering Shanghai Jiao Tong University China Key Laboratory of Shanghai Education Commission for Intelligent Interaction and Cognitive Engineering Shanghai Jiao Tong University China

Masked Language Modeling (MLM) has been widely used as the denoising objective in pretraining language models (PrLMs). Existing PrLMs commonly adopt a Random-Token Masking strategy where a fixed masking ratio is applied and different contents are masked by an equal probability throughout the entire training. However, the model may receive a complicated impact from pre-training status, which changes accordingly as training time goes on. In this paper, we show that such time-invariant MLM settings on masking ratio and masked content are unlikely to deliver an optimal outcome, which motivates us to explore the influence of time-variant MLM settings. We propose two scheduled masking approaches that adaptively tune the masking ratio and masked content in different training stages, which improves the pre-training efficiency and effectiveness verified on the downstream tasks. Our work is a pioneer study on time-variant masking strategy on ratio and content and gives a better understanding of how masking ratio and masked content influence the MLM pretraining1 Copyright © 2022, The Authors. All rights reserved.

关键词： Modeling languages

Security Enhancement in Consumer Enterprises using Neural Nets within the SIEM Framework

学校读者我要写书评

暂无评论

Security Enhancement in Consumer Enterprises using Neural Ne...

2025 IEEE International Conference on Consumer Electronics, ICCE 2025

作者： Arora, Saksham Kumar, Sudhakar Singh, Sunil K. Garg, Sahil Gupta, Brij B. Bansal, Shavi Chui, Kwok Tai Chandigarh College of Engineering and Technology Chandigarh India Department of Computer Science and Information Engineering Asia University Taiwan University of Economics and Human Sciences Warsaw Poland Insights2Techinfo India Hong Kong Metropolitan University Hong Kong

ISBN: (纸本)9798331521165

With internet popularity increasing every day and more physical activities being brought to the online spectrum, Security Information and Event Management (SIEM) tools have emerged as the main way to tackle any threat intrusions in the network infrastructure of organizations. Threats are evolving and becoming more advanced with the advancements in technology, and detecting and preventing the attacks need more human intervention than ever. To prevent these attacks, artificial intelligence and neural networks are a great way to tackle the increasing needs of human intervention. This article, thus, proposes a novel framework that uses a neural network named FeedForward to develop a model for anomaly and threat protection. It then prompts the user and takes input when a threat is detected. On confirmation of a threat, the feedback is sent back to the neural network to train itself better for future threats and keep itself updated. Therefore, this article proposes a supervised learning and retraining framework on the neural network. The dataset chosen has 763,144 instances that are divided into 80% training and 20% testing dataset. On running the neural network on the dataset, we achieved an astonishing 99.2% accuracy. © 2025 IEEE.

关键词： Feedforward neural networks

CGMatch: A Different Perspective of Semi-supervised Learning

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Cheng, Bo Lu, Jueqing Tian, Yuan Zhao, Haifeng Chang, Yi Du, Lan School of Artificial Intelligence Jilin University China International Center of Future Science Jilin University China Engineering Research Center of Knowledge-Driven Human-Machine Intelligence MOE China Faculty of Information Technology Monash University Australia Department of Computer Science Jinling Institute of Technology China

Semi-supervised learning (SSL) has garnered significant attention due to its ability to leverage limited labeled data and a large amount of unlabeled data to improve model generalization performance. Recent approaches achieve impressive successes by combining ideas from both consistency regularization and pseudo-labeling. However, these methods tend to underperform in the more realistic situations with relatively scarce labeled data. We argue that this issue arises because existing methods rely solely on the model’s confidence, making them challenging to accurately assess the model’s state and identify unlabeled examples contributing to the training phase when supervision information is limited, especially during the early stages of model training. In this paper, we propose a novel SSL model called CGMatch, which, for the first time, incorporates a new metric known as Count-Gap (CG). We demonstrate that CG is effective in discovering unlabeled examples beneficial for model training. Along with confidence, a commonly used metric in SSL, we propose a fine-grained dynamic selection (FDS) strategy. This strategy dynamically divides the unlabeled dataset into three subsets with different characteristics: easy-to-learn set, ambiguous set, and hard-to-learn set. By selective filtering subsets, and applying corresponding regularization with selected subsets, we mitigate the negative impact of incorrect pseudo-labels on model optimization and generalization. Extensive experimental results on several common SSL benchmarks indicate the effectiveness of CGMatch especially when the labeled data are particularly limited. Source code is available at https://***/BoCheng-96/CGMatch. Copyright © 2025, The Authors. All rights reserved.

关键词： Self-supervised learning