检索结果-内蒙古大学图书馆

Online knowledge distillation with elastic peer

INFORMATION SCIENCES 2022年 583卷 1-13页

作者： Tan, Chao Liu, Jie Natl Univ Def Technol Sci & Parallel & Distributed Proc Lab Changsha 410073 Peoples R China Natl Univ Def Technol Lab Software Engn Complex Syst Changsha 410073 Peoples R China

Knowledge distillation is a highly effective method for transferring knowledge from a cum-bersome teacher network to a lightweight student network. However, teacher networks are not always available. An alternative method called online knowledge distillation, which applies a group of peer networks to learn collaboratively with each other, has been pro -posed previously. In this study, we revisit online knowledge distillation and find that the existing training strategy limits the diversity among peer networks. Thus, online knowledge distillation cannot achieve its full potential. To address this, a novel online knowledge dis-tillation with elastic peer (KDEP) strategy is introduced here. The entire training process is divided into two phases by KDEP. In each phase, a specific training strategy is applied to adjust the diversity to an appropriate degree. Extensive experiments have been conducted on individual benchmarks, including CIFAR-100, CINIC-10, Tiny ImageNet, and Caltech-UCSD Birds. The results demonstrate the superiority of KDEP. For example, when the peer networks are ShuffleNetV2-1.0 and ShuffleNetV2-0.5, the target peer network ShuffleNetV2-0.5 achieves 57:00% top-1 accuracy on Tiny ImageNet via KDEP. (c) 2021 Elsevier Inc. All rights reserved.

关键词： Neural network compression Knowledge distillation Knowledge transfer

来源：评论

学校读者我要写书评

暂无评论

A generative adversarial network based on an efficient transformer for high-fidelity flow field reconstruction

引用

PHYSICS OF FLUIDS 2024年第7期36卷 075184-075184页

作者： Shen, Liming Deng, Liang Liu, Xuliang Wang, Yueqing Chen, Xinhai Liu, Jie Natl Univ Def Technol Lab Digitizing Software Frontier Equipment Changsha 410000 Peoples R China Natl Univ Def Technol Sci & Technol Parallel & Distributed Proc Lab Changsha 410000 Peoples R China China Aerodynam Res & Dev Ctr State Key Lab Aerodynam Mianyang 621000 Peoples R China China Aerodynam Res & Dev Ctr Computat Aerodynam Inst Mianyang 621000 Peoples R China

The reconstruction of high-fidelity flow fields from low-fidelity data has attracted considerable attention in fluid dynamics but poses many challenges to existing deep learning methods due to the spatiotemporal complexity of flows and the lack of standardized benchmark datasets. In this study, we generate a low- and high-fidelity dataset containing 25 600 snapshots of four representative flow dynamics simulations using eight different numerical-precision and grid-resolution configurations. Using this dataset, we develop a physics-guided transformer-based generative adversarial network (PgTransGAN) for concurrently handling numerical-precision and grid-resolution enhancement. PgTransGAN leverages a dual-discriminator-based generative adversarial network for capturing continuous spatial and temporal dynamics of flows and applies a soft-constraint approach to enforce physical consistency in the reconstructed data using gradient information. An efficient transformer model is also developed to obtain the long-term temporal dependencies and further alleviate storage constraints. We compare the performance of PgTransGAN against standard linear interpolation and solutions based solely on convolutional neural networks or generative adversarial networks, and demonstrate that our method achieves better reconstruction quality at the data, image, and physics levels with an upscaling factor of 4 or even 8 in each grid dimension.

关键词： Digital storage

来源：评论

学校读者我要写书评

暂无评论

Topic and Reference Guided keyphrase Generation from Social Media 1

引用

15th International Conference on Knowledge Science, Engineering, and Management (KSEM)

作者： Yu, Xiubin Chen, Xingjun Huang, Zhen Dou, Yong Hu, Biao Natl Univ Def Technol Natl Key Lab Parallel & Distributed Proc Changsha Peoples R China Dalian Navy Acad Dalian Peoples R China

ISBN: (数字)9783031109867

ISBN: (纸本)9783031109867;9783031109850

Automatic keyphrase generation can help human efficiently understand or process critical information from massive social media posts. Seq2Seq-based generation models that can produce both present and absent keyphrases have achieved remarkable performance. However, existing models are limited by the sparseness of posts that are widely exhibited in social media language. Sparseness makes it difficult for models to obtain useful features and generate accurate keyphrases. To address this problem, we propose the Topic and Reference Guided keyphrase Generation model(TRGKG) for social media posts, which enrich scarce posts features by corpus-level topics and post-level reference knowledge. The proposed model incorporates a contextual neural topic model to exploit topics and a heterogeneous graph to capture reference knowledge from retrieved related posts. To guide the decoding process, we introduce new topic-aware hierarchical attention and copy mechanism, which directly copies appropriate words from both the source post and its references. Experiments on two public datasets demonstrate that TRGKG achieves state-of-the-art performance.

关键词： keyphrase generation Seq2Seq Neural topic model Heterogeneous graph

来源：评论

学校读者我要写书评

暂无评论

Disentangled Orchestration on Cyber Ranges

引用

IEEE TRANSACTIONS ON DEPENDABLE AND SECURE COMPUTING 2024年第4期21卷 2344-2360页

作者： Fu, Yongquan Han, Weihong Yuan, Dong Natl Univ Def Technol Coll Comp Natl Key Lab Parallel & Distributed Comp Changsha 410073 Peoples R China Guangzhou Univ Sch Comp Sci & Cyber Engn Guangzhou 511370 Guangdong Peoples R China Univ Sydney Fac Engn Sydney NSW 2050 Australia

Cyber ranges require networked applications to test cyberspace events effectively. As testing becomes more advanced, it involves multiple real-world applications with flexible execution orders. However, it is increasingly challenging to orchestrate large-scale, chained, and heterogeneous Internet applications. State-of-the-art orchestration techniques face scalability issues due to inefficient representation models and entangled scheduling of events and applications. To address these issues, we present Wukong, a disentangled orchestration system in cyber ranges that disaggregates the scheduling and execution of workflows and their applications in a decentralized coordination approach. First, we overcome the heterogeneity of events with a workflow model that encodes event chains with compositional Directed Acyclic Graphs (DAGs) and unified event triggers. Second, Wukong disaggregates the execution of DAGs and applications with push-pull decentralized coordination over distributed agents. Our evaluation of Wukong on a real-world cyber range demonstrates its expressive, scalable, and efficient abilities for automatically emulating diverse event chains. The storage footprint of compositional modeling is up to 57 times smaller than that of baseline models. Wukong's response delay is 1.52 to 2.74 times shorter than state-of-the-art orchestration engines, and the scheduling delay is up to 2.16 times smaller than the baseline approach.

关键词： Task analysis Containers Behavioral sciences Delays Testing Chatbots Malware Composition event chaining experiment orchestration cyber range

来源：评论

学校读者我要写书评

暂无评论

F3A: Fairness-Aware AI-Workloads Allocation Considering Multidimensional User Demands in JointCloud 15

F3A: Fairness-Aware AI-Workloads Allocation Considering Mult...

引用

15th IEEE International Conference on Joint Cloud Computing (JCC)

作者： Yang, Jiacheng Yi, Guodong Gao, Fei Shi, Peichang Wang, Huaimin Natl Univ Def Technol Coll Comp Sci Natl Key Lab Parallel & Distributed Proc Changsha 410073 Peoples R China Xiangjiang Lab Changsha 410073 Peoples R China Hunan Univ Technol & Business Sch Adv Interdisciplinary Studies Changsha 410073 Peoples R China

ISBN: (纸本)9798350387339

With the rapid growth of large language models, cloud computing has become an indispensable component of the AI industry. Cloud service providers(CSPs) are establishing AI data centers to service AI workloads. In the face of this surging need for AI computing power, building a connected computing environment across various clouds and forming a JointCloud presents an attractive solution. However, scheduling AI tasks across multiple AI data centers within a JointCloud environment presents a significant challenge: how to balance users' demands while ensuring CSPs' fairness in scheduling. Existing research primarily focuses on optimizing scheduling quality with limited consideration for fairness. Therefore, this paper proposes a Fairness-Aware AI-Workloads Allocation method (F3A), a fair cross-cloud allocation technique for AI tasks. F3A utilizes Point and Token to reflect both the resource status and historical task allocations of AI data centers, enabling the consideration of users' multidimensional demands and facilitating fair task allocation across multiple centers. In order to better assess the fairness of scheduling, we also devised a fairness indicator(FI), based on the Gini coefficient to measure the fairness of task allocation. The experimental results demonstrate that F3A consistently maintains FI within 0.1 across various cluster sizes and different task quantities, representing an improvement of 76.45% compared to classical fair scheduling algorithms round-robin. F3A exhibits commendable performance in ensuring fairness in task allocation while also demonstrating effectiveness in cost reduction and enhancing user satisfaction.

关键词： JointCloud Computing Fairness Task Scheduling AI Workloads AI data center

来源：评论

学校读者我要写书评

暂无评论

Research on fast prediction of boiling flow parameters in rod bundle of NPP based on efficient CFD-ROM methods

引用

PROGRESS IN NUCLEAR ENERGY 2025年 184卷

作者： Mo, Jinhong Dong, Xiaomeng Xu, Yujie Xu, Anqi Yu, Yang Yang, Ming Shenzhen Univ China Inst Nucl Energy & Safety Coll Phys & Optoelect Engn Shenzhen Key Lab Nucl & Radiat Safety Nanhai St 3688 Nanshan Dist Shenzhen 518060 Peoples R China Natl Univ Def Technol Natl Key Lab Parallel & Distributed Comp Changsha 410073 Peoples R China Nucl Power Inst China Chengdu 610041 Peoples R China

Historically, Computational Fluid Dynamics (CFD) has been widely used to verify the flow dynamics in rod bundle channels. Nevertheless, the iterative calculation and time consumption make it impractical for the application of digital twin that require efficiency. In order to tackle this difficulty, we suggest utilizing a datadriven order reduced model (ROM) to rapidly predict flow fields of rod bundle channels. Previous research on Reduced Order Models (ROM) for rod bundle channel has primarily focused on the single-phase state, with little emphasis on the two-phase state. Thus, the study focuses on the ROM of rod bundle channel in a two-phase state. First, a Computational Fluid Dynamics (CFD) data set is processed using the Proper Orthogonal Decomposition (POD) algorithm to identify important modes. Then, a Back-Propagation Neural Network (BPNN) model is trained as the agent model, due to its strong ability to fit non-linear relationship between input and output. The structure of BPNN is optimized. Ultimately, the ROM model is utilized to predict the flow field of a rod bundle channel under new boundary conditions. The comparison with CFD calculation confirms the effectiveness of the Reduced Order Model (ROM), showcasing its high precision in predicting the temperature, velocity, and void fraction fields within the rod bundle channel. In addition, the ROM model reaches a computation speed that is roughly 10<^>4 times quicker than standard CFD simulations. Nonetheless, it is noted that the model's predictive accuracy diminishes under conditions of slight subcooled boiling, due to the lack of samples under the condition. An improvement strategy: Subdivide the sampling space region according to various change stages, implement partition sampling.

关键词： POD algorithm BPNN model Boiling flow CFD numerical simulation Rod bundle channel model

来源：评论

学校读者我要写书评

暂无评论

DaMSTF: Domain Adversarial Learning Enhanced Meta Self-Training for Domain Adaptation 61

DaMSTF: Domain Adversarial Learning Enhanced Meta Self-Train...

引用

61st Annual Meeting of the the Association-for-Computational-Linguistics (ACL)

作者： Lu, Menglong Huang, Zhen Zhao, Yunxiang Tian, Zhiliang Liu, Yang Li, Dongsheng Natl Univ Def Technol Natl Key Lab Parallel & Distributed Comp Changsha Peoples R China Beijing Inst Biotechnol Beijing Peoples R China

ISBN: (纸本)9781959429722

Self-training emerges as an important research line on domain adaptation. By taking the model's prediction as the pseudo labels of the unlabeled data, self-training bootstraps the model with pseudo instances in the target domain. However, the prediction errors of pseudo labels (label noise) challenge the performance of self-training. To address this problem, previous approaches only use reliable pseudo instances, i.e., pseudo instances with high prediction confidence, to retrain the model. Although these strategies effectively reduce the label noise, they are prone to miss the hard examples. In this paper, we propose a new self-training framework for domain adaptation, namely Domain adversarial learning enhanced Self-Training Framework (DaMSTF). Firstly, DaMSTF involves meta-learning to estimate the importance of each pseudo instance, so as to simultaneously reduce the label noise and preserve hard examples. Secondly, we design a meta constructor for constructing the meta validation set, which guarantees the effectiveness of the meta-learning module by improving the quality of the meta validation set. Thirdly, we find that the meta-learning module suffers from the training guidance vanishment and tends to converge to an inferior optimal. To this end, we employ domain adversarial learning as a heuristic neural network initialization method, which can help the meta-learning module converge to a better optimal. Theoretically and experimentally, we demonstrate the effectiveness of the proposed DaMSTF. On the cross-domain sentiment classification task, DaMSTF improves the performance of BERT with an average of nearly 4%.

关键词： Heuristic methods

来源：评论

学校读者我要写书评

暂无评论

SAC: An Ultra-Efficient Spin-based Architecture for Compressed DNNs

引用

ACM TRANSACTIONS ON ARCHITECTURE AND CODE OPTIMIZATION 2024年第1期21卷 1-26页

作者： Zhao, Yunping Ma, Sheng Liu, Heng Huang, Libo Dai, Yi Inst Microelect & Microprocessors Sch Computy Changsha Peoples R China Natl Univ Def Technol Sci & Technol Parallel & Distributed Proc Lab Changsha Peoples R China Natl Univ Def Technol Sch Comp Changsha Peoples R China Natl Univ Def Technol Inst Microelect & Microprocessors Sch Comp Changsha Peoples R China

Deep Neural Networks (DNNs) have achieved great progress in academia and industry. But they have become computational and memory intensive with the increase of network depth. Previous designs seek breakthroughs in software and hardware levels to mitigate these challenges. At the software level, neural network compression techniques have effectively reduced network scale and energy consumption. However, the conventional compression algorithm is complex and energy intensive. At the hardware level, the improvements in the semiconductor process have effectively reduced power and energy consumption. However, it is difficult for the traditional Von-Neumann architecture to further reduce the power consumption, due to the memory wall and the end of Moore's law. To overcome these challenges, the spintronic device based DNN machines have emerged for their non-volatility, ultra low power, and high energy efficiency. However, there is no spinbased design that has achieved innovation at both the software and hardware level. Specifically, there is no systematic study of spin-based DNN architecture to deploy compressed networks. In our study, we present an ultra-efficient Spin-based Architecture for Compressed DNNs (SAC), to substantially reduce power consumption and energy consumption. Specifically, we propose a One-Step Compression algorithm (OSC) to reduce the computational complexity with minimum accuracy loss. We also propose a spin-based architecture to realize better performance for the compressed network. Furthermore, we introduce a novel computation flow that enables the reuse of activations and weights. Experimental results show that our study can reduce the computational complexity of compression algorithm from O(Tk(3)) to O(k(2) log k), and achieve 14x similar to 40x compression ratio. Furthermore, our design can attain a 2x enhancement in power efficiency and a 5x improvement in computational efficiency compared to the Eyeriss. Our models are available at an anonymous

关键词： Artificial intelligence chips Neural network hardware Neuromorphic computing

来源：评论

学校读者我要写书评

暂无评论

PCSAGAN: a physics-constrained generative network based on self-attention for high-fidelity flow field reconstruction

引用

JOURNAL OF VISUALIZATION 2024年第4期27卷 661-676页

作者： Shen, Liming Deng, Liang Wang, Yueqing Zhang, Jian Liu, Jie Natl Univ Def Technol Sci & Technol Parallel & Distributed Proc Lab Changsha Peoples R China Natl Univ Def Technol Lab Digitizing Software Frontier Equipment Changsha Peoples R China China Aerodynam Res & Dev Ctr Computat Aerodynam Inst Mianyang Peoples R China China Aerodynam Res & Dev Ctr State Key Lab Aerodynam Mianyang Peoples R China

We propose a physics-constrained generative adversarial network, PCSAGAN, based on the self-attention mechanism for high-fidelity flow field reconstruction, which can generate high-resolution and high-precision volumes for two-dimensional time-varying datasets. PCSAGAN consists of two discriminators and one generator, capable of preserving temporal coherence and spatial characteristics. It avoids the heavy computation of multilayer convolution and obtains spatial flow evolution information by using the self-attention mechanism. A soft-constraint loss function is further employed to effectively utilize the underlying physical properties. Moreover, the dataset is generated from real-world simulations rather than artificial synthesis, which ensures generalizability to practical applications. We compare PCSAGAN against volume upscaling methods using BI, ESPCN and SSR-TVD, and demonstrate its effectiveness with several time-varying flow fields through quantitative and qualitative evaluations. Meanwhile, a re-analysis framework is introduced to enable potential later visualization and exploratory analysis of the reconstructed flow field data.

关键词： Time-varying data visualization Deep learning Super-resolution Generative adversarial network

来源：评论

学校读者我要写书评

暂无评论

GGN: a model-free, data-driven deep learning framework for reconstructing gene regulatory networks 23

GGN: a model-free, data-driven deep learning framework for r...

引用

13th International Conference on Bioscience, Biochemistry and Bioinformatics (ICBBB)

作者： Mao, Guo Pang, Zhengbin Zuo, Ke Liu, Jie Natl Univ Def Technol Sci & Technol Parallel & Distributed Proc Lab Changsha Peoples R China Natl Univ Def Technol Lab Software Engn Complex Syst Changsha Peoples R China

ISBN: (纸本)9781450398190

Reconstructing gene regulatory networks based on time-series gene expression data is a huge challenge in the field of systems biology. However, the accuracy of traditional methods can be further improved. In practical situations, the gene regulatory network topology and the dynamic rules of genes are not observable. Therefore, reconstructing the underlying network structure and dynamics from observed time-series gene expression data is an important task. In this work, we introduce a new framework, Gumbel Graph Network (GGN), a model-free, data-driven deep learning framework for accomplishing the reconstruction of gene regulatory networks and the reconstruction of gene dynamics. Our model consists of two co-trained parts: a network generator, which generates a discrete network using Gumbel Softmax technique;and a dynamic learner, which uses the generated network and single-step trajectory values to predict the state at the next moment. According to the experimental results on multiple simulated and real datasets, GGN has better performance and stability compared with other state-of-the-art algorithms.

关键词： Gumbel Graph Network Network Generator Gumbel Softmax Technique Gynamic Learner Gene Regulatory Network

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：