检索结果-内蒙古大学图书馆

arXiv 2024年

作者： Huang, Nianchang Yang, Yang Zhang, Qiang Han, Jungong Huang, Jin The State Key Laboratory of Electromechanical Integrated Manufacturing of High-Performance Electronic Equipments The Center for Complex Systems School of Mechano-Electronic Engineering Xidian University Shaanxi Xi’an710071 China The Department of Computer Science University of Sheffield United Kingdom The State Key Laboratory of Electromechanical Integrated Manufacturing of High-Performance Electronic Equipment Shaanxi Xi’an710071 China

This paper delves into the task of arbitrary modality salient object detection (AM SOD), aiming to detect salient objects from arbitrary modalities, e.g. RGB images, RGB-D images, and RGB-D-T images. A novel modality-adaptive Transformer (MAT) will be proposed to investigate two fundamental challenges of AM SOD, i.e. more diverse modality discrepancies caused by varying modality types that need to be processed, and dynamic fusion design caused by an uncertain number of modalities present in the inputs of multimodal fusion strategy. Specifically, inspired by prompt learning’s ability of aligning the distributions of pre-trained models to the characteristic of downstream tasks by learning some prompts, MAT will first present a modality-adaptive feature extractor (MAFE) to tackle the diverse modality discrepancies by introducing a modality prompt for each modality. In the training stage, a new modality translation contractive (MTC) loss will be further designed to assist MAFE in learning those modality-distinguishable modality prompts. Accordingly, in the testing stage, MAFE can employ those learned modality prompts to adaptively adjust its feature space according to the characteristics of the input modalities, thus being able to extract discriminative unimodal features. Then, MAFE will present a channel-wise and spatial-wise fusion hybrid (CSFH) strategy to meet the demand for dynamic fusion. For that, CSFH dedicates a channel-wise dynamic fusion module (CDFM) and a novel spatial-wise dynamic fusion module (SDFM) to fuse the unimodal features from varying numbers of modalities and meanwhile effectively capture cross-modal complementary semantic and detail information, respectively. Moreover, CSFH will carefully align CDFM and SDFM to different levels of unimodal features based on their characteristics for more effective complementary information exploitation. Experimental results show that by virtue of MAFE, MTC loss and CSFH, our proposed MAT achieves significant inc

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

Vision-Amplified Semantic Entropy for Hallucination Detection in Medical Visual Question Answering

arXiv

引用

arXiv 2025年

作者： Liao, Zehui Hu, Shishuai Zou, Ke Fu, Huazhu Zhen, Liangli Xia, Yong National Engineering Laboratory for Integrated Aero-Space-Ground-Ocean Big Data Application Technology School of Computer Science and Engineering Northwestern Polytechnical University Xi’an710072 China Institute of High Performance Computing Agency for Science Technology and Research Singapore138632 Singapore National University of Singapore Singapore119077 Singapore

Multimodal large language models (MLLMs) have demonstrated significant potential in medical Visual Question Answering (VQA). Yet, they remain prone to hallucinations—incorrect responses that contradict input images, posing substantial risks in clinical decision-making. Detecting these hallucinations is essential for establishing trust in MLLMs among clinicians and patients, thereby enabling their real-world adoption. Current hallucination detection methods, especially semantic entropy (SE), have demonstrated promising hallucination detection capacity for LLMs. However, adapting SE to medical MLLMs by incorporating visual perturbations presents a dilemma. Weak perturbations preserve image content and ensure clinical validity, but may be overlooked by medical MLLMs, which tend to over rely on language priors. In contrast, strong perturbations can distort essential diagnostic features, compromising clinical interpretation. To address this issue, we propose Vision Amplified Semantic Entropy (VASE), which incorporates weak image transformations and amplifies the impact of visual input, to improve hallucination detection in medical VQA. We first estimate the semantic predictive distribution under weak visual transformations to preserve clinical validity, and then amplify visual influence by contrasting this distribution with that derived from a distorted image. The entropy of the resulting distribution is estimated as VASE. Experiments on two medical open-ended VQA datasets demonstrate that VASE consistently outperforms existing hallucination detection methods. Copyright © 2025, The Authors. All rights reserved.

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

Low Latency Conversion of Artificial Neural Network Models to Rate-encoded Spiking Neural Networks

arXiv

引用

arXiv 2022年

作者： Yan, Zhanglu Zhou, Jun Wong, Weng-Fai Department of Computer Science National University of Singapore Singapore117417 Singapore Agency for Science Technology and Research Institute of High Performance Computing Singapore138632 Singapore

Spiking neural networks (SNNs) are well suited for resource-constrained applications as they do not need expensive multipliers. In a typical rate-encoded SNN, a series of binary spikes within a globally fixed time window is used to fire the neurons. The maximum number of spikes in this time window is also the latency of the network in performing a single inference, as well as determines the overall energy efficiency of the model. The aim of this paper is to reduce this while maintaining accuracy when converting ANNs to their equivalent SNNs. The state-of-the-art conversion schemes yield SNNs with accuracies comparable with ANNs only for large window sizes. In this paper, we start with understanding the information loss when converting from pre-existing ANN models to standard rate-encoded SNN models. From these insights, we propose a suite of novel techniques that together mitigate the information lost in the conversion, and achieve state-of-art SNN accuracies along with very low latency. Our method achieved a Top-1 SNN accuracy of 98.73% (1 time step) on the MNIST dataset, 76.38% (8 time steps) on the CIFAR-100 dataset, and 93.71% (8 time steps) on the CIFAR-10 dataset. On ImageNet, an SNN accuracy of 75.35%/79.16% was achieved with 100/200 time steps. Copyright © 2022, The Authors. All rights reserved.

关键词： Neural networks

来源：评论

学校读者我要写书评

暂无评论

Ggf: Global Geometric Feature for Rotation Invariant Point Cloud Understanding

SSRN

引用

SSRN 2022年

作者： Xiao, Yunzhe Li, Fu Wang, Haotian Yang, Shaowu Institute for Quantum Information State Key Laboratory of High Performance Computing College of Computer Science and Technology National University of Defense Technology Changsha410073 China

Some point cloud understanding tasks are supposed to be rotation-invariant. However, rotation-invariance is not promised in most existing deep learning approaches. Extra data argumentation or a complex professional network is always required to increase the generalization to arbitrary orientations, which cannot benefit generic point cloud understanding tasks. To avoid these problems, we propose a simple yet generic plug-and-play feature, namely Global Geometric Feature(GGF), to eliminate the rotation influence in generic point cloud understanding tasks. GGF projects the point cloud to a rotation-invariant feature space by extracting the geometric relationship among points and obtaining a distributed global description. We find that GGF can be easily plugged into the generic point cloud processing approaches as a low-level feature without further adapting. The extensive experimental results show that the proposed GGF can enable the generic framework to achieve comparable performance accuracy compared to the dedicated rotation-invariant methods. © 2022, The Authors. All rights reserved.

关键词： Rotation

来源：评论

学校读者我要写书评

暂无评论

Interference-Aware Opportunistic Job Placement for Shared Distributed Deep Learning Clusters

SSRN

引用

SSRN 2023年

作者： Li, Hongliang Zhao, Hairui Sun, Ting Li, Xiang Xu, Haixiao Li, Keqin College of Computer Science and Technology Jilin University Changchun China Key Laboratory of Symbolic Computation and Knowledge Engineering The Ministry of Education Changchun China High Performance Computing Center Jilin University Changchun China Department of Computer Science State University of New York New PaltzNY12561 United States

Distributed deep learning frameworks facilitate large deep learning workloads. These frameworks support sharing one GPU device among multiple jobs to improve resource utilization. Modern deep learning training jobs consume a large amount of GPU memory. Despite that, sharing GPU memory among jobs is still possible because a training job has iterative steps that its memory usage fluctuates over time. However, resource sharing also introduces the risk of job performance degradation. Colocated jobs sharing a GPU device may suffer from different levels of interference, mainly caused by memory oversharing. How to improve resource utilization while maintaining good job performance is a novel challenge for job placement strategies. This paper studies the job placement problem. We propose an opportunistic memory sharing model to describe the time-varying job memory requirements. Based on this model, we introduce an Opportunistic Job Placement Problem (OJPP) for shared GPU clusters that seek job placement configurations using a minimum number of GPU devices and guarantee user-defined performance requirements at the same time. We propose a greedy algorithm and a heuristic algorithm with computational complexities of O(n log n) and O(n2 log n), respectively, to solve the problem. We also propose an online adjustment algorithm with the computational complexity of O(n log n) to perform updates to job placement configurations in runtime. A machine-learning-based interference prediction method is used to prepare accurate interference estimations. Extensive experiments are conducted on a GPU cluster to verify the correctness and effectiveness of our algorithms. Compared with standalone training jobs on dedicated clusters, the proposed approach reduces resource consumption by 46% in a shared cluster, while guaranteeing over 92.97% of the job performance, in terms of average job completion time. © 2023, The Authors. All rights reserved.

关键词： Computational complexity

来源：评论

学校读者我要写书评

暂无评论

An Unsupervised Short- and Long-Term Mask Representation for Multivariate Time Series Anomaly Detection

arXiv

引用

arXiv 2022年

作者： Miao, Qiucheng Xu, Chuanfu Zhan, Jun Zhu, Dong Wu, Chengkun Institute for Quantum Information State Key Laboratory of High Performance Computing College of Computer Science and Technology National University of Defense Technology Hunan Changsha China

Anomaly detection of multivariate time series is meaningful for system behavior monitoring. This paper proposes an anomaly detection method based on unsupervised Short- and Long-term Mask Representation learning(SLMR). The main idea is to extract short-term local dependency patterns and long-term global trend patterns by using multi-scale residual dilated convolution and Gated Recurrent Unit(GRU) respectively. Furthermore, our approach can comprehend temporal contexts and feature correlations by combining spatial-temporal masked self-supervised representation learning and sequence split. It considers the importance of features is different, and we introduce the attention mechanism to adjust the contribution of each feature. Finally, a forecasting-based model and a reconstruction-based model are integrated to focus on single timestamp prediction and latent representation of time series. Experiments show that the performance of our method outperforms other state-of-the-art models on three real-world datasets. Further analysis shows that our method is good at anomaly localization. © 2022, CC BY.

关键词： Convolution

来源：评论

学校读者我要写书评

暂无评论

Topic-Guided Self-Introduction Generation for Social Media Users

arXiv

引用

arXiv 2023年

作者： Xu, Chunpu Li, Jing Li, Piji Yang, Min Department of Computing The Hong Kong Polytechnic University Hong Kong College of Computer Science and Technology Nanjing University of Aeronautics and Astronautics China Shenzhen Key Laboratory for High Performance Data Mining Shenzhen Institutes of Advanced Technology Chinese Academy of Sciences China

Millions of users are active on social media. To allow users to better showcase themselves and network with others, we explore the auto-generation of social media self-introduction, a short sentence outlining a user’s personal interests. While most prior work profiles users with tags (e.g., ages), we investigate sentence-level self-introductions to provide a more natural and engaging way for users to know each other. Here we exploit a user’s tweeting history to generate their self-introduction. The task is non-trivial because the history content may be lengthy, noisy, and exhibit various personal interests. To address this challenge, we propose a novel unified topic-guided encoder-decoder (UTGED) framework;it models latent topics to reflect salient user interest, whose topic mixture then guides encoding a user’s history and topic words control decoding their self-introduction. For experiments, we collect a large-scale Twitter dataset, and extensive results show the superiority of our UTGED to the advanced encoder-decoder models without topic modeling. 1 © 2023, CC BY.

关键词： User profile

来源：评论

学校读者我要写书评

暂无评论

A First-Order Multi-Gradient Algorithm for Multi-Objective Bi-Level Optimization

arXiv

引用

arXiv 2024年

作者： Ye, Feiyang Lin, Baijiong Cao, Xiaofeng Zhang, Yu Tsang, Ivor W. Department of Computer Science and Engineering Southern University of Science and Technology China Australian Artificial Intelligence Institute University of Technology Sydney Australia China School of Artificial Intelligence Jilin University China Shanghai Artificial Intelligence Laboratory China Centre for Frontier AI Research Agency for Science Technology and Research Singapore Institute of High Performance Computing Agency for Science Technology and Research School of Computer Science and Engineering Nanyang Technological University Singapore

In this paper, we study the Multi-Objective Bi-Level Optimization (MOBLO) problem, where the upper-level subproblem is a multi-objective optimization problem and the lower-level subproblem is for scalar optimization. Existing gradient-based MOBLO algorithms need to compute the Hessian matrix, causing the computational inefficient problem. To address this, we propose an efficient first-order multi-gradient method for MOBLO, called FORUM. Specifically, we reformulate MOBLO problems as a constrained multi-objective optimization (MOO) problem via the value-function approach. Then we propose a novel multi-gradient aggregation method to solve the challenging constrained MOO problem. Theoretically, we provide the complexity analysis to show the efficiency of the proposed method and a non-asymptotic convergence result. Empirically, extensive experiments demonstrate the effectiveness and efficiency of the proposed FORUM method in different learning problems. In particular, it achieves state-of-the-art performance on three multi-task learning benchmark datasets. The code is available at https://***/Baijiong-Lin/FORUM. Copyright © 2024, The Authors. All rights reserved.

关键词： Optimization algorithms

来源：评论

学校读者我要写书评

暂无评论

Building Device Models Using LLMs for Requirements Simulation in Smart Homes

Building Device Models Using LLMs for Requirements Simulatio...

引用

IEEE International Requirements Engineering Conference Workshops (REW)

作者： Shi Chen Xiaohong Chen Zhi Jin Shanghai Key Laboratory of Trustworthy Computing East China Normal University Shanghai China Key Laboratory of High-Confidence Software Technologies Peking University Ministry of Education China Department of Computer Science and Technology Peking University Beijing China

ISBN: (数字)9798350395518

ISBN: (纸本)9798350395525

The implementation of model-based requirements simulations in smart homes is an effective strategy that allows users to comprehend the execution impact of their personalized needs before deployment. This necessitates the use of device models in advance. In this paper, we propose an automatic modeling method for smart home devices that utilizes a Large Language Model (LLM) to assist end users in modeling their devices. The method involves extracting information directly from device manuals using the extraction capabilities of the LLM, deducing the automata models using the reasoning capabilities of the LLM, and verifying the model quality through the properties of valid device models. Our evaluation has demonstrated that our method can generate high-quality device models. The resultant model can be utilized for smart home scenario simulations, assisting users in validating their requirements. Thus, this provides a cost-effective and scalable solution for offering device models to simulate user requirements in smart homes.

关键词： Large language models Conferences Automata Smart homes Manuals Information retrieval Cognition

来源：评论

学校读者我要写书评

暂无评论

FlyDB: Query Optimization of Blockchain System Based on Hybrid Storage Architecture

FlyDB: Query Optimization of Blockchain System Based on Hybr...

引用

Artificial Intelligence, Networking and Information Technology (AINIT), International Seminar on

作者： Yang Liu Zehui Xie Nong Xiao Zhiguang Chen Yang Ou Institute for Quantum Information & State Key Laboratory of High-Performance Computing College of Computer National University of Defense Technology Changsha China School of computer science and Engineering Sun Yat-sen University Guangzhou China College of Computer National University of Defense Technology Changsha China

The hybrid storage architecture incorporates a blockchain dual-index structure optimized for byte addressing in non-volatile memory, low-latency features, and the large capacity of SSD. This paper presents a novel LSM-Tree architecture that leverages the byte-addressing characteristics of non-volatile memory. The layers of the level 0 and level 1 are stored in non-volatile memory while employing a local key-value separation strategy to minimize read and write operations during compaction. A logic compression strategy is proposed to handle data compression tasks efficiently. A dual-index structure design is introduced for users to address the scope and multi-keyword queries in the blockchain system. During the compression process, keyword switching occurs during data reordering, ensuring that data within the same SSTable is rearranged according to user preferences. By effectively harnessing the advantages of each level in the hybrid storage architecture, this design significantly reduces both databases write time and the response time for complex queries.

关键词：

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：