检索结果-内蒙古大学图书馆

Power Optimization for Low Transmission Delay in Software Defined Data Center Networks

IEEE Transactions on Mobile computing 2025年第7期24卷 5716-5730页

作者： Zhao, Yong Cao, Jiannong Wang, Xingwei Li, Fuliang He, Qiang Liu, Xiaojie University of Electronic Science and Technology of China Trusted Cloud Computing and Big Data Key Laboratory of Sichuan Province Laboratory of Intelligent Collaborative Computing Chengdu611731 China Hong Kong Polytechnic University Department of Computing Hung Hom Hong Kong Northeastern University College of Computer Science and Engineering Shenyang110819 China Northeastern University College of Medicine and Biological Information Engineering Shenyang110819 China Pengcheng Laboratory Shenzhen518066 China

Software-Defined Data Center Networks (SDDCNs) utilizes Software Defined Networking (SDN) as a network architecture to achieve highly flexible, programmable, and automated management of Data Center Networks (DCNs). The high energy consumption of DCNs remains a persistent and significant challenge. Thus, the energy saving is crucial and imperative for DCNs. Current energy-efficient solutions primarily rely on flow consolidation and dynamic device sleeping techniques to reduce energy consumption. However, these approaches often yield long Flow Completion Time (FCT), potentially resulting in violations of service-level agreements, particularly for delay-sensitive applications. In this paper, we formulate the problem of minimizing the power consumption in SDDCNs as key objective while ensuring timely FCT for delay-sensitive applications. To solve this problem, we first introduce the Active Network Generation (ANG) approach, which generates a minimal active subnet with the least number of active devices while meeting the current traffic demand. Subsequently, we propose two algorithms based on the type of applications: the Delay-Tolerant Flow Route (DTFR) algorithm for delay-tolerant applications and the Delay-Sensitive Flow Route (DSFR) algorithm for delay-sensitive applications. Simulation results demonstrate that our propose solution achieves an energy-saving rate of up to 67.77% and significantly reduces FCT compared to benchmark solutions. © 2002-2012 IEEE.

关键词： Energy efficiency

来源：评论

学校读者我要写书评

暂无评论

Llama SLayer 8B: Shallow Layers Hold the Key to Knowledge Injection

Llama SLayer 8B: Shallow Layers Hold the Key to Knowledge In...

引用

2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024

作者： Chen, Tianxiang Tan, Zhentao Gong, Tao Wu, Yue Chu, Qi Liu, Bin Ye, Jieping Yu, Nenghai School of Cyber Science and Technology University of Science and Technology of China China Anhui Province Key Laboratory of Digital Security China CAS Key Laboratory of Electromagnetic Space Information China Alibaba Cloud China

ISBN: (纸本)9798891761681

As a manner to augment pre-trained large language models (LLM), knowledge injection is critical to develop vertical domain large models and has been widely studied. Although most current approaches, including parameter-efficient fine-tuning (PEFT) and block expansion methods, uniformly apply knowledge across all LLM layers, it raises the question: are all layers equally crucial for knowledge injection? We begin by evaluating the importance of each layer in finding the optimal layer range for knowledge injection. Intuitively, the more important layers should play a more critical role in knowledge injection and deserve a denser injection. We observe performance dips in question-answering benchmarks after the removal or expansion of the shallow layers, and the degradation shrinks as the layer gets deeper, indicating that the shallow layers hold the key to knowledge injection. This insight leads us to propose the S strategy, a post-pretraining strategy of selectively enhancing shallow layers while pruning the less effective deep ones. Based on this strategy, we introduce Llama Slayer-8B and Llama Slayer-8B-Instruct. We experimented on the corpus of code & math and demonstrated the effectiveness of our strategy. Further experiments across different LLM, Mistral-7B, and a legal corpus confirmed the general applicability of the approach, underscoring its wide-ranging efficacy. Our code is available at: https://***/txchenUSTC/Llama-Slayer. © 2024 Association for Computational Linguistics.

关键词： Computational linguistics

来源：评论

学校读者我要写书评

暂无评论

Speech Translation Refinement using Large Language Models

arXiv

引用

arXiv 2025年

作者： Dou, Huaixia Tian, Xinyu Lyu, Xinglin Zhu, Jie Li, Junhui Guo, Lifan School of Computer Science and Technology Soochow University Suzhou China Alibaba Cloud Computing Hangzhou China

Recent advancements in large language models (LLMs) have demonstrated their remarkable capabilities across various language tasks. Inspired by the success of text-to-text translation refinement, this paper investigates how LLMs can improve the performance of speech translation by introducing a joint refinement process. Through the joint refinement of speech translation (ST) and automatic speech recognition (ASR) transcription via LLMs, the performance of the ST model is significantly improved in both training-free in-context learning and parameter-efficient fine-tuning scenarios. Additionally, we explore the effect of document-level context on refinement under the context-aware fine-tuning scenario. Experimental results on the MuST-C and CoVoST 2 datasets, which include seven translation tasks, demonstrate the effectiveness of the proposed approach using several popular LLMs including GPT-3.5-turbo, LLaMA3-8B, and Mistral-12B. Further analysis further suggests that jointly refining both transcription and translation yields better performance compared to refining translation alone. Meanwhile, incorporating document-level context significantly enhances refinement performance. We release our code and datasets on GitHub1 Copyright © 2025, The Authors. All rights reserved.

关键词： Speech enhancement

来源：评论

学校读者我要写书评

暂无评论

ControlMath: Controllable Data Generation Promotes Math Generalist Models

ControlMath: Controllable Data Generation Promotes Math Gene...

引用

2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024

作者： Chen, Nuo Wu, Ning Chang, Jianhui Shou, Linjun Li, Jia Hong Kong University of Science and Technology Hong Kong Microsoft United States China Telecom Cloud Computing Research Institute China

ISBN: (纸本)9798891761643

Utilizing large language models (LLMs) for data augmentation has yielded encouraging results in mathematical reasoning. However, these approaches face constraints in problem diversity, potentially restricting them to in-domain/distribution data generation. To this end, we propose ControlMath, an iterative method involving an equation-generator module and two LLM-based agents. The module creates diverse equations, which the Problem-Crafter agent then transforms into math word problems. The Reverse-Agent filters and selects high-quality data, adhering to the "less is more" principle, achieving better results with fewer data points. This approach enables the generation of diverse math problems, not limited to specific domains or distributions. As a result, we collect ControlMathQA, which involves 190k math word problems. Extensive results prove that combining our dataset with in-domain datasets like GSM8K can help improve the model's mathematical ability to generalize, leading to improved performances both within and beyond specific domains. © 2024 Association for Computational Linguistics.

关键词： Computational linguistics

来源：评论

学校读者我要写书评

暂无评论

Incremental Transformer: Efficient Encoder for Incremented Text Over MRC and Conversation Tasks 31

Incremental Transformer: Efficient Encoder for Incremented T...

引用

31st International Conference on Computational Linguistics, COLING 2025

作者： Li, Weisheng Wang, Yuechen Shi, Jiaxin Zhou, Wengang Li, Houqiang Tian, Qi University of Science and Technology of China China Institute of Artificial Intelligence Hefei Comprehensive National Science Center China Huawei Cloud Computing Technologies Co. Ltd. China

ISBN: (纸本)9798891761964

Some encoder inputs such as conversation histories are frequently extended with short additional inputs like new responses. However, to obtain the real-time encoding of the extended input, existing Transformer-based encoders like BERT have to encode the whole extended input again without utilizing the existing encoding of the original input, which may be prohibitively slow for real-time applications. In this paper, we introduce Incremental Transformer, an efficient encoder dedicated for faster encoding of incremented input. It takes only added input as input but attends to cached representations of original input in lower layers for better performance. By treating questions as additional inputs of a passage, Incremental Transformer can also be applied to accelerate MRC tasks. Experimental results show tiny decline in effectiveness but significant speedup against traditional full encoder across various MRC and multi-turn conversational question answering tasks. With the help from simple distillation-like auxiliary losses, Incremental Transformer achieves a speedup of 6.2x, with a mere 2.2 point accuracy reduction in comparison to RoBERTa-Large on SQuADV1.1. © 2025 Association for Computational Linguistics.

关键词： Computational linguistics

来源：评论

学校读者我要写书评

暂无评论

SimAC: A Simple Anti-Customization Method for Protecting Face Privacy Against Text-to-Image Synthesis of Diffusion Models

SimAC: A Simple Anti-Customization Method for Protecting Fac...

引用

Conference on Computer Vision and Pattern Recognition (CVPR)

作者： Feifei Wang Zhentao Tan Tianyi Wei Yue Wu Qidong Huang University of Science and Technology of China Alibaba Cloud

ISBN: (数字)9798350353006

ISBN: (纸本)9798350353013

Despite the success of diffusion-based customization methods on visual content creation, increasing concerns have been raised about such techniques from both privacy and political perspectives. To tackle this issue, several anti-customization methods have been proposed in very recent months, predominantly grounded in adversarial attacks. Unfortunately, most of these methods adopt straight-forward designs, such as end-to-end optimization with afocus on adversarially maximizing the original training loss, thereby neglecting nuanced internal properties intrinsic to the diffusion model, and even leading to ineffective optimization in some diffusion time steps. In this paper, we strive to bridge this gap by undertaking a comprehensive exploration of these inherent properties, to boost the performance of current anti-customization approaches. Two aspects of properties are investigated: 1) We examine the relationship between time step selection and the model's perception in the frequency domain of images and find that lower time steps can give much more contributions to adversarial noises. This inspires us to propose an adaptive greedy search for optimal time steps that seamlessly integrates with existing anti-customization methods. 2) We scrutinize the roles of features at different layers during denoising and devise a sophisticated feature-based optimization framework for anti-customization. Experiments on facial benchmarks demonstrate that our approach significantly increases identity disruption, thereby protecting user privacy and copyright. Our code is available at:: http//***/somuchtome/SimAC.

关键词： Training Privacy Adaptation models Visualization Frequency-domain analysis Noise reduction Text to image

来源：评论

学校读者我要写书评

暂无评论

Less is More: Improving LLM Alignment via Preference Data Selection

arXiv

引用

arXiv 2025年

作者： Deng, Xun Zhong, Han Ai, Rui Feng, Fuli Wang, Zheng He, Xiangnan University of Science and Technology of China China Peking University China Massachusetts Institute of Technology United States Alibaba Cloud Computing China

Direct Preference Optimization (DPO) has emerged as a promising approach for aligning large language models with human preferences. While prior work mainly extends DPO from the aspect of the objective function, we instead improve DPO from the largely overlooked but critical aspect of data selection. Specifically, we address the issue of parameter shrinkage caused by noisy data by proposing a novel margin-maximization principle for dataset curation in DPO training. To accurately estimate margins for data selection, we propose a dual-margin guided approach that considers both external reward margins and implicit DPO reward margins. Extensive experiments demonstrate that our method reduces computational cost dramatically while improving performance. Remarkably, by using just 10% of the Ultrafeedback dataset, our approach achieves 3% to 8% improvements across various Llama and Mistral series models on the AlpacaEval 2.0 benchmark. Furthermore, our approach seamlessly extends to iterative DPO, yielding a roughly 3% improvement with 25% online data, while further reducing training time. These results highlight the potential of data selection strategies for advancing preference optimization. Copyright © 2025, The Authors. All rights reserved.

关键词： Data assimilation

来源：评论

学校读者我要写书评

暂无评论

Research on Used Car Price Prediction Based on Topological Data Analysis and Attention Convolutional Neural Networks 24

Research on Used Car Price Prediction Based on Topological D...

引用

2024 Guangdong-Hong Kong-Macao Greater Bay Area International Conference on Digital Economy and Artificial Intelligence, DEAI 2024

作者： Tang, Ying Tan, Yuan Guangdong University of Science and Technology Guangdong Dongguan523000 China Huawei Cloud Computing Technology Co. Ltd. Guangdong Dongguan523000 China

ISBN: (纸本)9798400717147

With the continuous growth of the national economy, car ownership in china has been steadily increasing, leading to a continuous expansion of the used car market. The formulation of prices for used cars is a crucial factor that affects and restricts the healthy development of the used car market. Therefore, this study addresses the pricing issue in the used car transactions market and proposes a pricing assessment model-the TACNN (Topological Data Analysis and Attention Convolutional Neural Networks) model. The model ingeniously combines spatio-temporal features with topological features applied to regression prediction, allowing for both point prediction and interval prediction simultaneously. Comparative experiments on the kaggle used car price prediction competition dataset show that the TACNN model outperforms the baseline model in terms of accuracy and robustness. The relevant research findings are of significant importance in guiding the rational formulation of prices in used car transactions and promoting the healthy and orderly development of the used car market. © 2024 ACM.

关键词： Neural network models

来源：评论

学校读者我要写书评

暂无评论

DON’T TAKE THINGS OUT OF CONTEXT: ATTENTION INTERVENTION FOR ENHANCING CHAIN-OF-THOUGHT REASONING IN LARGE LANGUAGE MODELS

arXiv

引用

arXiv 2025年

作者： Yan, Shaotian Shen, Chen Wang, Wenxiao Xie, Liang Liu, Junjie Ye, Jieping Alibaba Cloud Computing China College of Software Zhejiang University China Zhejiang University of Technology College of Computer Science and Technology China

Few-shot Chain-of-Thought (CoT) significantly enhances the reasoning capabilities of large language models (LLMs), functioning as a whole to guide these models in generating reasoning steps toward final answers. However, we observe that isolated segments, words, or tokens within CoT demonstrations can unexpectedly disrupt the generation process of LLMs. The model may overly concentrate on certain local information present in the demonstration, introducing irrelevant noise into the reasoning process and potentially leading to incorrect answers. In this paper, we investigate the underlying mechanism of CoT through dynamically tracing and manipulating the inner workings of LLMs at each output step, which demonstrates that tokens exhibiting specific attention characteristics are more likely to induce the model to take things out of context;these tokens directly attend to the hidden states tied with prediction, without substantial integration of non-local information. Building upon these insights, we propose a Few-shot Attention Intervention method (FAI) that dynamically analyzes the attention patterns of demonstrations to accurately identify these tokens and subsequently make targeted adjustments to the attention weights to effectively suppress their distracting effect on LLMs. Comprehensive experiments across multiple benchmarks demonstrate consistent improvements over baseline methods, with a remarkable 5.91% improvement on the AQuA dataset, further highlighting the effectiveness of FAI. © 2025, CC BY.

关键词： Chains

来源：评论

学校读者我要写书评

暂无评论

TSM-LLM: Task Scheduling Management System for Large Language Models 5

TSM-LLM: Task Scheduling Management System for Large Languag...

引用

5th International Conference on Big Data and Artificial Intelligence and Software Engineering, ICBASE 2024

作者： Wen, Ziqiang Zhu, Guoping Wang, Yong Luo, Haijun Liu, Nianchao Beijing University of Posts and Telecommunications School of Computer Science Beijing China China Electronics Cloud Computing Technology Co. Ltd Cloud Product R&d Department Wuhan China

ISBN: (纸本)9798331506612

As large model services gain popularity;the high cost of deployment presents a challenge. Thus, focusing on improving model efficiency in high-load scenarios while managing the scheduling of multiple models. This paper introduces TSMLLM, a Task Scheduling Management System designed for optimizing the deployment of large language models (LLMs). The proposed system leverages an adaptive time window to group realtime requests, optimizing the batching process for better resource utilization. A two-step scheduling method, involving filtering and scoring, is employed to tailor task distribution based on the specific characteristics of each large model. The system's scalable architecture supports seamless expansion to accommodate evolving model requirements. Experimental results demonstrate a significant increase in throughput with only a slight rise in response time, contributing to a reduction in overall service costs. © 2024 IEEE.

关键词： Batch data processing

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：