检索结果-内蒙古大学图书馆

Adaptive Pruning for Large Language Models with Structural Importance Awareness

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zheng, Haotian Ren, Jinke Sun, Yushan Zhang, Ruichen Zhang, Wenbo Li, Zhen Niyato, Dusit Cui, Shuguang Han, Yatong National Key Laboratory of Autonomous Marine Vehicle Technology Harbin Engineering University Harbin150001 China The Chinese University of Hong Kong Shenzhen518172 China The Guangdong Provincial Key Laboratory of Future Networks of Intelligence The Chinese University of Hong Kong Shenzhen518172 China College of Computing and Data Science Nanyang Technological University Singapore Co. Ltd Shenzhen518048 China SSE The FNii-Shenzhen The Guangdong Provincial Key Laboratory of Future Networks of Intelligence The Chinese University of Hong Kong Shenzhen518172 China FNii-Shenzhen The Guangdong Provincial Key Laboratory of Future Networks of Intelligence The Chinese University of Hong Kong Shenzhen518172 China Infused Synapse AI Shenzhen518048 China

The recent advancements in large language models (LLMs) have significantly improved language understanding and generation capabilities. However, it is difficult to deploy LLMs on resource-constrained edge devices due to their high computational and storage resource demands. To address this issue, we propose a novel LLM model pruning method, namely structurally-aware adaptive pruning (SAAP), to significantly reduce the computational and memory costs while maintaining model performance. We first define an adaptive importance fusion metric to evaluate the importance of all coupled structures in LLMs by considering their homoscedastic uncertainty. Then, we rank the importance of all modules to determine the specific layers that should be pruned to meet particular performance requirements. Furthermore, we develop a new group fine-tuning strategy to improve the inference efficiency of LLMs. Finally, we evaluate the proposed SAAP method on multiple LLMs across two common tasks, i.e., zero-shot classification and text generation. Experimental results show that our SAAP method outperforms several state-of-the-art baseline methods, achieving 2.17%, 2.37%, and 2.39% accuracy gains on LLaMA-7B, Vicuna- 7B, and LLaMA-13B. Additionally, SAAP improves the token generation speed by 5%, showcasing its practical advantages in resource-constrained scenarios. Copyright © 2024, The Authors. All rights reserved.

关键词： Classification (of information)

Cross Modal Sentiment Classification of Social Media Based on Meta Heuristic Algorithm

学校读者我要写书评

暂无评论

Cross Modal Sentiment Classification of Social Media Based o...

International Conference on Computer Supported Cooperative Work in Design

作者： Wantong Du Wanli Min Yushui Geng Hu Liang Hehu Zhou Ministry of Education Shandong Computer Science Center Key Laboratory of Computing Power Network and Information Security Qilu University of Technology (Shandong Academy of Sciences) Jinan China Shandong Engineering Research Center of Big Data Applied Technology Faculty of Computer Science and Technology Qilu University of Technology (Shandong Academy of Sciences) Jinan China Shandong Provincial Key Laboratory of Computer Networks Shandong Fundamental Research Center for Computer Science Jinan China Synthesis Electronic Technology Co. Ltd

ISBN: (数字)9798350349184

ISBN: (纸本)9798350349191

The evolution of online social media has reshaped public behaviors, and social data infused with emotions provides crucial decision support for sentiment analysis tasks. Conventional multimodal approaches, influenced by redundant information in feature extraction during sentiment analysis, often focus solely on modality interactions without considering the unique content of each modality. This paper introduces the HGB-HFN model, addressing feature redundancy by optimizing extracted features using metaheuristic algorithm techniques, achieving a higher accuracy optimal feature dataset at lower computational costs. To counteract the neglect of modal-independent information, a hybrid fusion approach is adopted. It prioritizes visual features to extract precise semantic and emotional information from textual content, training multiple base classifiers to learn independent and diverse discriminative information. Comparative experiments on streaming datasets MOSI and MOSEI demonstrate the computational approach’s superiority in both accuracy and efficiency over existing methods.

关键词： Training Sentiment analysis visualization Accuracy Social networking (online) Metaheuristics Semantics

GenFace: A Large-Scale Fine-Grained Face Forgery Benchmark and Cross Appearance-Edge Learning

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zhang, Yaning Yu, Zitong Wang, Tianyi Huang, Xiaobin Shen, Linlin Gao, Zan Ren, Jianfeng Computer Vision Institute College of Computer Science and Software Engineering Shenzhen University Shenzhen518060 China School of Computing and Information Technology Great Bay University Dongguan523000 China National Engineering Laboratory for Big Data System Computing Technology Shenzhen University Shenzhen518060 China Nanyang Technological University 50 Nanyang Ave Block N 4 639798 Singapore Shenzhen Institute of Artificial Intelligence and Robotics for Society Shenzhen518129 China Guangdong Key Laboratory of Intelligent Information Processing Shenzhen University China Jinan250014 China Key Laboratory of Computer Vision and System Ministry of Education Tianjin University of Technology Tianjin300384 China School of Computer Science University of Nottingham Ningbo China

The rapid advancement of photorealistic generators has reached a critical juncture where the discrepancy between authentic and manipulated images is increasingly indistinguishable. Thus, benchmarking and advancing techniques detecting digital manipulation become an urgent issue. Although there have been a number of publicly available face forgery datasets, the forgery faces are mostly generated using GAN-based synthesis technology, which does not involve the most recent technologies like diffusion. The diversity and quality of images generated by diffusion models have been significantly improved and thus a much more challenging face forgery dataset shall be used to evaluate SOTA forgery detection literature. In this paper, we propose a large-scale, diverse, and fine-grained high-fidelity dataset, namely GenFace, to facilitate the advancement of deepfake detection, which contains a large number of forgery faces generated by advanced generators such as the diffusion-based model and more detailed labels about the manipulation approaches and adopted generators. In addition to evaluating SOTA approaches on our benchmark, we design an innovative Cross Appearance-Edge Learning (CAEL) detector to capture multi-grained appearance and edge global representations, and detect discriminative and general forgery traces. Moreover, we devise an Appearance-Edge Cross-Attention (AECA) module to explore the various integrations across two domains. Extensive experiment results and visualizations show that our detection model outperforms the state of the arts on different settings like cross-generator, cross-forgery, and cross-dataset evaluations. Code and datasets will be available at https://***/Jenine-321/GenFace. Copyright © 2024, The Authors. All rights reserved.

关键词： Image enhancement

Deep Learning Strategies for Addressing Anomalous Exposure in Image Processing: The FARDBUNet Approach 5

学校读者我要写书评

暂无评论

Deep Learning Strategies for Addressing Anomalous Exposure i...

5th International Conference on High Performance Big data and Intelligent Systems, HDIS 2023

作者： Zhou, Qi Yang, Kai Ke, Zunwang Wang, Gang Zhang, Yugui Jia, Yizhen Cao, Fengcai Ma, Junxiao Liu, Changlin Zhang, Kaijie Wu, Min School of Semiconductor Science and Technology South China Normal University GuangZhou510631 China CGNPC Uranium Industry Development Co. Ltd. Beijing100020 China School of Software Xinjiang University Urumqi830046 China School of Computing and Data Engineering NingboTech University Ningbo315100 China AnnLab Institute of Semiconductors Chinese Academy of Sciences Beijing100083 China Beijing Key Laboratory of Semiconductor Neural Network Intelligent Sensing and Computing Technology Beijing100083 China School of Integrated Circuits University of Chinese Academy of Sciences Beijing100049 China Co. Ltd Beijing100176 China School of Information Beijing Forestry University Beijing100083 China

ISBN: (纸本)9798350394160

In real-world scenarios, capturing scenes with excessive dynamic range often leads to the partial loss of highlight or dark area information due to irradiance variations and limitations in the capture capabilities of imaging devices caused by their size and material constraints. This phenomenon gives rise to anomalous exposure problems and triggers the loss of crucial initial details, thereby becoming a significant obstacle to achieving high image quality. Despite numerous researchers’ integration of exposure compensation techniques into the imaging process, the anomalous exposure problem remains unsolved, hindering progress in image processing. Consequently, considerable research is valuable in implementing exposure calibration techniques on visual perception and multimedia devices. This dissertation proposes FARDBUNet, a novel end-to-end exposure calibration model designed to address the anomalous exposure problem. A key innovation of FARDBUNet is the incorporation of a fused attention mechanism, enabling the model to prioritize regions needing enhanced correction while accounting for the global exposure characteristics of the image. Additionally, to tackle the challenge of inadequate information availability in the localized areas due to exposure issues, we introduce an atrous residual dense block that efficiently captures local information by expanding the receptive field. Comprehensive tests validate the effectiveness of our suggested model, which performs on par with leading techniques on expansive exposure datasets. ©2023 IEEE.

关键词： Image enhancement

Larger receptive Field and Context Information for Pose estimation:Larger Gaussian kernel 5

学校读者我要写书评

暂无评论

Larger receptive Field and Context Information for Pose esti...

5th International Conference on High Performance Big data and Intelligent Systems, HDIS 2023

作者： Ma, Junxiao Yang, Kai Ke, Zunwang Wang, Gang Zhang, Yugui Cao, Fengcai Zhou, Qi Fei, Yang School of Information Beijing Forestry University Beijing100083 China CGNPC Uranium Industry Development Co. Ltd. Beijing100020 China School of Software Xinjiang University Urumqi830046 China School of Computing and Data Engineering NingboTech University Ningbo315100 China AnnLab Institute of Semiconductors Chinese Academy of Sciences Beijing100083 China Beijing Key Laboratory of Semiconductor Neural Network Intelligent Sensing and Computing Technology Beijing100083 China School of Integrated Circuits University of Chinese Academy of Sciences Beijing100049 China Co. Ltd Beijing100176 China School of Semiconductor Science and Technology South China Normal University GuangZhou510631 China

ISBN: (纸本)9798350394160

The field of pose estimation has a wide range of application prospects in various industries in the current era. With the continuous development of deep learning techniques, the effects in the field of human pose estimation are constantly being optimized. However, due to the influence of factors such as occlusion and human posture variability, it is easy to lead to inaccurate identification of keypoints and even keypoints’ missing and matching errors, there is still a lot of optimization space for the effect of pose estimation. In this paper, we propose an improvement to generate heatmaps in the training process: increase the radius of the Gaussian kernel and adjust the smoothness of Gaussian distribution to get a larger receptive field, encourage the model to use more context information in prediction, and control other parameters unchanged on the premise of using heatmap regression. Improve keypoint detection performance by optimizing this item. In the experiment, for the COCO-keypoint data set, we selected different methods published in recent years, and after our improvement, we obtained better accuracy and performance than the original method, which proves the validity of our method. ©2023 IEEE.

关键词： Gaussian distribution

学校读者我要写书评

暂无评论

Proceedings of the Fortieth Conference on Uncertainty in Artificial Intelligence

作者： Zhiqiang Zhang Yongqiang Jiang Qian Gao Zhipeng Wang Key Laboratory of Computing Power Network and information Security Ministry of Eaucation Shandong Computer Science Center Qilu University of Technology (Shandong Academy of Sciences) Jinan China and Shandong Engineering Research Center of Big Data Applied Technology Faculty of Computer Science and Technology Qilu University of Technology (Shandong Academy of Sciences) Jinan China and Shandong Provincial Key Laboratory of Computer Networks Shandong Fundamental Research Center for Computer Science Jinan China China Telecom Digital Intelligence Techonology Co Jinan China

Existing multimodal recommendation research still faces some challenges, such as not being able to fully mine the implicit relevance information of neighbor nodes, and the unreasonable weight allocation to imbalanced nodes. To address the aforementioned challenges, this paper introduces a new multimodal recommendation model called NS-MAR+. Specifically, the model firstly constructs a neighbor similarity graph convolutional network to capture the implicit relevance information and reasonably assigns the attention weights through the graph attention mechanism. Secondly, the model introduces a modal alignment and fusion mechanism by using a multilayer perceptron (MLP) to map image and text features into a shared space for comparison and fusion. In addition, the model constructs a user co-interaction graph and an item semantic graph based on the original information and performs graph convolution operations to enhance the preference information of users and items, and to better capture the interactions and internal features between users and items. Finally, MLP is employed to aggregate user and item representations and predict personalized recommendation rankings. To validate the experiment's efficacy, this paper compares with several leading multimodal recommendation models on public datasets with a performance improvement of 1% to 3%. The experimental outcomes indicate that the model in this paper has good superiority and accuracy in multimodal recommendation tasks.

关键词：

JC5A: Service Delay Minimization for Aerial MEC-assisted Industrial Cyber-Physical Systems

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Sun, Geng Wu, Jiaxu Sun, Zemin He, Long Wang, Jiacheng Niyato, Dusit Jamalipour, Abbas Mao, Shiwen College of Computer Science and Technology Jilin University Changchun130012 China Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University Changchun130012 China College of Computing and Data Science Nanyang Technological University Singapore639798 Singapore School of Computer Science and Engineering Nanyang Technological University Singapore639798 Singapore School of Electrical and Computer Engineering The University of Sydney SydneyNSW2006 Australia Department of Electrical and Computer Engineering Auburn University Auburn United States

In the era of the sixth generation (6G) and industrial Internet of Things (IIoT), an industrial cyber-physical system (ICPS) drives the proliferation of sensor devices and computing-intensive tasks. To address the limited resources of IIoT sensor devices, unmanned aerial vehicle (UAV)-assisted mobile edge computing (MEC) has emerged as a promising solution, providing flexible and cost-effective services in close proximity of IIoT sensor devices (ISDs). However, leveraging aerial MEC to meet the delay-sensitive and computation-intensive requirements of the ISDs could face several challenges, including the limited communication, computation and caching (3C) resources, stringent offloading requirements for 3C services, and constrained on-board energy of UAVs. To address these issues, we first present a collaborative aerial MEC-assisted ICPS architecture by incorporating the computing capabilities of the macro base station (MBS) and UAVs. We then formulate a service delay minimization optimization problem (SDMOP). Since the SDMOP is proved to be an NP-hard problem, we propose a joint computation offloading, caching, communication resource allocation, computation resource allocation, and UAV trajectory control approach (JC5A). Specifically, JC5A consists of a block successive upper bound minimization method of multipliers (BSUMM) for computation offloading and service caching, a convex optimization-based method for communication and computation resource allocation, and a successive convex approximation (SCA)-based method for UAV trajectory control. Moreover, we theoretically prove the convergence and polynomial complexity of JC5A. Simulation results demonstrate that the proposed approach can achieve superior system performance compared to the benchmark approaches and algorithms. Copyright © 2024, The Authors. All rights reserved.

关键词： Convex optimization

Remaining Useful Life Prediction with Spatio-Temporal Graph Transform and Weakly Supervised Adversarial Network: An Application in Power Components

学校读者我要写书评

暂无评论

SSRN

SSRN 2024年

作者： Deng, Shuhan Lan, Hao Yue, Ke Chen, Zhuyun Zhang, Xuning Huang, Ruyi Huang, Zhicong Li, Weihua Shien-Ming Wu School of Intelligent Engineering South China University of Technology Guangzhou China School of Mechanical & Automotive Engineering South China University of Technology Guangzhou China State Key Laboratory of Precision Electronic Manufacturing Technology and Equipment Guangdong University of Technology Guangzhou China Department of Industrial Engineering and Decision Analytics The Hong Kong University of Science and Technology Hong Kong College of Computing and Data Science Nanyang Technological University Singapore Singapore

Predictive maintenance is crucial for improving the reliability of power components and having attracted widespread attention, for which remaining useful life (RUL) prediction is a precondition. Owing to the heterogeneity of the power components, accuracy RUL prediction through different components remains challenging. To make things worse, the structural property of the data itself is also neglected. Herein, take the insulated gate bipolar transistors (IGBTs) as the sample, a novel RUL prediction framework is proposed in this paper. In brief, a time window and Radius distance metric function is adopted to generate graph information matrix. Specifically, spatio-temporal fusion graph network (STFGN) is proposed to obtain unique IGBTs data structure information and incorporate past and future moments aging features to construct health index (HI). Then, weakly supervised adversarial learning strategy is considered to realize the cross-domain RUL prediction, which conditional domain adaptation (CDA)-based adversarial method is adopted to reduce distribution discrepancies between distinct domains and weakly supervised training strategy with some target domain labels is also used to enhance positive transfer. Moreover, the proposed method is proved by four transfer tasks and the results show that overcomes the challenge of unavailability of labeled failure samples and outperforms state-of-the-art methods. © 2024, The Authors. All rights reserved.

关键词： Forecasting

A Comprehensive Survey on Underwater Image Enhancement Based on Deep Learning

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Cong, Xiaofeng Zhao, Yu Gui, Jie Hou, Junming Tao, Dacheng The School of Cyber Science and Engineering Southeast University Nanjing210000 China The School of Cyber Science and Engineering Southeast University Purple Mountain Laboratories Nanjing China Ministry of Education 210000 China The State Key Laboratory of Millimeter Waves School of Information Science and Engineering Southeast University Nanjing210096 China The College of Computing & Data Science Nanyang Technological University #32 Block N4 #02a-014 50 Nanyang Avenue Singapore639798 Singapore

Underwater image enhancement (UIE) presents a significant challenge within computer vision research. Despite the development of numerous UIE algorithms, a thorough and systematic review is still absent. To foster future advancements, we provide a detailed overview of the UIE task from several perspectives. Firstly, we introduce the physical models, data construction processes, evaluation metrics, and loss functions. Secondly, we categorize and discuss recent algorithms based on their contributions, considering six aspects: network architecture, learning strategy, learning stage, auxiliary tasks, domain perspective, and disentanglement fusion. Thirdly, due to the varying experimental setups in the existing literature, a comprehensive and unbiased comparison is currently unavailable. To address this, we perform both quantitative and qualitative evaluations of state-of-the-art algorithms across multiple benchmark datasets. Lastly, we identify key areas for future research in UIE. A collection of resources for UIE can be found at https://***/YuZhao1999/UIE. Copyright © 2024, The Authors. All rights reserved.

关键词： Network architecture