检索结果-内蒙古大学图书馆

41st International Conference on Machine Learning, ICML 2024

作者： Liu, Zirui Yuan, Jiayi Jin, Hongye Zhong, Shaochen Xu, Zhaozhuo Braverman, Vladimir Chen, Beidi Hu, Xia Department of Computer Science Rice University United States Department of Computer Science Texas A&M University United States Department of Computer Science Stevens Institute of Technology United States Department of Electrical and Computer Engineering Carnegie Mellon University United States

Efficiently serving large language models (LLMs) requires batching many requests together to reduce the cost per request. Yet, the key-value (KV) cache, which stores attention keys and values to avoid re-computations, significantly increases memory demands and becomes the new bottleneck in speed and memory usage. This memory demand increases with larger batch sizes and longer context lengths. Additionally, the inference speed is limited by the size of KV cache, as the GPU's SRAM must load the entire KV cache from the main GPU memory for each token generated, causing the computational core to be idle during this process. A straightforward and effective solution to reduce KV cache size is quantization, which decreases the total bytes taken by KV cache. However, there is a lack of in-depth studies that explore the element distribution of KV cache to understand the hardness and limitation of KV cache quantization. To fill the gap, we conducted a comprehensive study on the element distribution in KV cache of popular LLMs. Our findings indicate that the key cache should be quantized per-channel, i.e., group elements along the channel dimension and quantize them together. In contrast, the value cache should be quantized per-token. From this analysis, we developed a tuning-free 2bit KV cache quantization algorithm, named KIVI. With the hardware-friendly implementation, KIVI can enable Llama (Llama-2), Falcon, and Mistral models to maintain almost the same quality while using 2.6× less peak memory usage (including the model weight). This reduction in memory usage enables up to 4× larger batch size, bringing 2.35× ∼ 3.47× throughput on real LLM inference workload. The source code is available at https://***/jy-yuan/KIVI. Copyright 2024 by the author(s)

关键词： Cache memory

来源：评论

学校读者我要写书评

暂无评论

Heat Transfer Area Optimization for Heat Exchanger System

引用

computer Modeling in engineering & sciences 2025年第4期143卷 335-349页

作者： Yu-Cheng Liao Fu-I Chou Po-Yuan Yang Jyh-Horng Chou Department of Electrical Engineering National Kaohsiung University of Science and TechnologyKaohsiung807Taiwan Department of Intelligent Robotics National Pingtung UniversityPingtung900Taiwan Department of Mechanical and Computer-Aided Engineering Feng Chia UniversityTaichung407Taiwan

This paper presents an allowable-tolerance-based group search optimization(AT-GSO),which combines the robust GSO(R-GSO)and the external quality design planning of the Taguchi ***-GSO algorithm is used to optimize the heat transfer area of the heat exchanger *** R-GSO algorithm integrates the GSO algorithm with the Taguchi method,utilizing the Taguchi method to determine the optimal producer in each iteration of the GSO algorithm to strengthen the robustness of the search process and the ability to find the global *** conventional parameter design optimization,it is typically assumed that the designed parameters can be applied accurately and consistently throughout ***,for systems that are sensitive to changes in design parameters,even minor inaccuracies can substantially reduce overall system ***,the permissible variations of the design parameters are considered in the tolerance-optimized design to ensure the robustness of the *** optimized design of the heat exchanger system assumes that the system’s operating temperature parameters are ***,fixing the systemoperating temperature parameters at a constant value is *** paper assumes that the system operating temperature parameters have an uncertainty error when optimizing the heat transfer area of the heat exchanger *** results show that the AT-GSO algorithm optimizes the heat exchanger system and finds the optimal operating temperature in the absence of tolerance and under three tolerance conditions.

关键词： Heat exchanger system allowable tolerance-based group search optimization Taguchi method tolerance design

来源：评论

学校读者我要写书评

暂无评论

Feedback-based safe gradient flow for optimal regulation of virtual power plants 14

Feedback-based safe gradient flow for optimal regulation of ...

引用

14th Mediterranean Conference on Power Generation Transmission, Distribution and Energy Conversion, MEDPOWER 2024

作者： Colot, Antonin Cornélusse, Bertrand Cortés, Jorge Dall'Anese, Emiliano Department of Electrical Engineering and Computer Science University of Liège Belgium Department of Mechanical and Aerospace Engineering University of California San Diego United States Department of Electrical and Computer Engineering Boston University United States

ISBN: (纸本)9781837242689

This paper considers the problem of controlling distributed energy resources (DERs) in a distribution network (DN);the paper focuses on the voltage regulation task and on the concept of virtual power plant (VPP). For the latter, we envision an aggregation of DERs as a VPP that tracks power setpoints at the point of common coupling to provide ancillary services to the bulk power system. We propose a feedback-based controller that pursues solutions to a time-varying AC optimal power flow problem. This controller ensures voltage constraints are met in real time, assuming no voltage measurement errors. The feedback from the system mitigates the controller sensitivity to model uncertainties and eliminates the need for measurements of uncontrollable loads at every node. The feedback-based controller is tested on the IEEE37-node system with delta and wye-connected devices in a strongly unbalanced configuration. © The Institution of engineering & Technology 2024.

关键词： Virtual Power Plants

来源：评论

学校读者我要写书评

暂无评论

Augmented and End-to-End Models for Defect Classification of Structures 18

Augmented and End-to-End Models for Defect Classification of...

引用

18th ACM International Conference on Distributed and Event-Based Systems, DEBS 2024

作者： Kim, Gyu-Il Chung, Kyungyong Department of Computer Science Kyonggi University Suwon Korea Republic of Division of AI Computer Science and Engineering Kyonggi University Suwon Korea Republic of

ISBN: (纸本)9798400704437

This paper proposes an automated defect classification model for the safety diagnosis and maintenance of concrete structures. Traditional manual classification methods are time-consuming and rely on subjective judgment, whereas the proposed model utilizes time-series data obtained from sensors to enable faster and more accurate defect detection through data augmentation and Transformer-based classification. Data augmentation compensates for the imbalance in defect data distribution, and the Transformer effectively interprets the complex temporal patterns of the data, capturing long-range dependencies. This end-to-end model performs an integrated process from data processing to defect classification. Future research will explore the model's generalizability and the potential for integration into real-time monitoring systems. This paper presents a new approach to improving the economic and efficient evaluation of the safety of concrete structures. © 2024 Owner/Author.

关键词： Defects

来源：评论

学校读者我要写书评

暂无评论

Dynamic Defense Policies Learning Through Soft Actor-Critic for Soccer Robot Goalkeepers

Dynamic Defense Policies Learning Through Soft Actor-Critic ...

引用

2024 International Conference on Fuzzy Theory and Its Applications, iFUZZY 2024

作者： Li, Shih-An Huang, Chun-Jui Wong, Ching-Chang Feng, Hsuan-Ming Tamkang University Department of Electrical and Computer Engineering New Taipei City Taiwan National Quemoy University Department of Computer Science and Information Engineering Kinmen Taiwan

ISBN: (纸本)9798350352788

This paper implements a dynamic goalkeeper defense policy based on Deep Reinforcement Learning (DRL) for a soccer robot. This study proposed a multi-directional training approach to enable soccer goalkeepers to learn defense strategies for acting an appropriate manner. This paper using the Soft Actor-Critic (SAC) as DRL architecture and the Gazebo simulator as an environment simulation platform. Utilizing progressive learning with good reward function design to enable soccer robot goalkeepers to explore and learn in a simulated environment. It efficiently completed the better defense task. Finally, the correctness of the defense policy is evaluated by testing and validating it in a multi-directional area. The experimental results demonstrated the designed SAC containing the superior performance for successful defense rates in this paper. © 2024 IEEE.

关键词： Deep reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Handling Class Imbalance in Online Transaction Fraud Detection

引用

computers, Materials & Continua 2022年第2期70卷 2861-2877页

作者： Kanika Jimmy Singla Ali Kashif Bashir Yunyoung Nam Najam UI Hasan Usman Tariq School of Computer Science and Engineering Lovely Professional UniversityPunjabIndia Department of Computing and Mathematics ManchesterMetropolitan UniversityUK&School of Electrical Engineering and Computer ScienceNational University of Science and TechnologyIslamabadPakistan Department of Computer Science and Engineering Soonchunhyang UniversityAsan31538Korea Department of Electrical and Computer Engineering Dhofar UniversitySalalahOman College of Computer Engineering and Sciences Prince Sattam bin Abdulaziz UniversityAl-Kharj11942Saudi Arabia

With the rise of internet facilities,a greater number of people have started doing online transactions at an exponential rate in recent years as the online transaction system has eliminated the need of going to the bank physically for every ***,the fraud cases have also increased causing the loss of money to the ***,an effective fraud detection system is the need of the hour which can detect fraudulent transactions automatically in ***,the genuine transactions are large in number than the fraudulent transactions which leads to the class imbalance *** this research work,an online transaction fraud detection system using deep learning has been proposed which can handle class imbalance problem by applying algorithm-level methods which modify the learning of the model to focus more on the minority class i.e.,fraud transactions.A novel loss function named Weighted Hard-Reduced Focal Loss(WH-RFL)has been proposed which has achieved maximum fraud detection rate i.e.,True PositiveRate(TPR)at the cost of misclassification of few genuine transactions as high TPR is preferred over a high True Negative Rate(TNR)in fraud detection system and same has been demonstrated using three publicly available imbalanced transactional ***,Thresholding has been applied to optimize the decision threshold using cross-validation to detect maximum number of frauds and it has been demonstrated by the experimental results that the selection of the right thresholding method with deep learning yields better results.

关键词： Class imbalance deep learning fraud detection loss function thresholding

来源：评论

学校读者我要写书评

暂无评论

Reinforcement Learning-based Receding Horizon Control using Adaptive Control Barrier Functions for Safety-Critical Systems 63

Reinforcement Learning-based Receding Horizon Control using ...

引用

63rd IEEE Conference on Decision and Control, CDC 2024

作者： Sabouni, Ehsan Sabbir Ahmad, H.M. Giammarino, Vittorio Cassandras, Christos G. Paschalidis, Ioannis Ch. Li, Wenchao Boston University Division of Systems Engineering BostonMA02215 United States Boston University Department of Electrical and Computer Engineering The Division of Systems Engineering BostonMA02215 United States Boston University Department of Electrical and Computer Engineering Division of Systems Engineering Department of Biomedical Engineering BostonMA02215 United States

ISBN: (纸本)9798350316339

Optimal control methods provide solutions to safety-critical problems but easily become intractable. Control Barrier Functions (CBFs) have emerged as a popular technique that facilitates their solution by provably guaranteeing safety, through their forward invariance property, at the expense of some performance loss. This approach involves defining a performance objective alongside CBF-based safety constraints that must always be enforced with both performance and solution feasibility significantly impacted by two key factors: (i) the selection of the cost function and associated parameters, and (ii) the calibration of parameters within the CBF-based constraints, which capture the trade-off between performance and conservativeness. To address these challenges, we propose a Reinforcement Learning (RL)-based Receding Horizon Control (RHC) approach leveraging Model Predictive Control (MPC) with CBFs (MPC-CBF). In particular, we parameterize our controller and use bilevel optimization, where RL is used to learn the optimal parameters while MPC computes the optimal control input. We validate our method by applying it to the challenging automated merging control problem for Connected and Automated Vehicles (CAVs) at conflicting roadways. Results demonstrate improved performance and a significant reduction in the number of infeasible cases compared to traditional heuristic approaches used for tuning CBF-based controllers, showcasing the effectiveness of the proposed method. In order to guarantee reproducibility, our code is provided here: ***/EhsanSabouni/CDC2024_RL_adpative_MPC_CBF/tree/main © 2024 IEEE.

关键词： Adaptive control systems

来源：评论

学校读者我要写书评

暂无评论

Continuous Query-based Data Trading

引用

Performance Evaluation Review 2024年第1期52卷 73-74页

作者： Cheng, Jin Ding, Ningning Lui, John C.S. Huang, Jianwei School of Science and Engineering Shenzhen Institute of Artificial Intelligence and Robotics for Society The Chinese University of Hong Kong Shenzhen Shenzhen China Department of Electrical and Computer Engineering Northwestern University Evanston United States Department of Computer Science and Engineering The Chinese University of Hong Kong Hong Kong Hong Kong

In the era of big data, traditional data trading methods designed for one-time queries on static databases fail to meet the demands of continuous query-based trading on streaming data, often resulting in repeated and inaccurate charges due to neglecting computation sharing in continuous query execution. To address this, we introduce CQTrade, the first trading framework tailored for continuous queries, which incorporates computation sharing across different time windows and queries, enhancing integration with existing trading systems. Our contributions include a theoretical analysis of computation-sharing techniques, the development of a general optimization problem to maximize seller revenue adaptable to various techniques, and the proposal of a branch-and-price algorithm to handle the problem's complexity. Our evaluation shows that the proposed framework improves the success rate of data trading by 12.8% and boosts the seller's revenue by an average of 28.7%, compared to the one-time query-based data trading methods used in the current data market. © 2024 Owner/Author.

关键词： Integer programming

来源：评论

学校读者我要写书评

暂无评论

Smart Devices Based Multisensory Approach for Complex Human Activity Recognition

引用

computers, Materials & Continua 2022年第2期70卷 3221-3234页

作者： Muhammad Atif Hanif Tallha Akram Aamir Shahzad Muhammad Attique Khan Usman Tariq Jung-In Choi Yunyoung Nam Zanib Zulfiqar Department of Electrical and Computer Engineering COMSATS University IslamabadWah Campus47040Pakistan Department of Electrical and Computer Engineering COMSATS University IslamabadAbbottabad Campus22040Pakistan Department of Computer Science HITEC University TaxilaTaxilaPakistan College of Computer Engineering and Sciences Prince Sattam Bin Abdulaziz UniversityAl-KhrajSaudi Arabia Department of Applied Artificial Intelligence Ajou UniversitySuwonKorea Department of Computer Science and Engineering Soonchunhyang UniversityAsanKorea Department of Electrical and Computer Engineering CUISahiwalPakistan

Sensors based Human Activity Recognition(HAR)have numerous applications in eHeath,sports,fitness assessments,ambient assisted living(AAL),human-computer interaction and many *** human physical activity can be monitored by using wearable sensors or external *** usage of external devices has disadvantages in terms of cost,hardware installation,storage,computational time and lighting conditions ***,most of the researchers used smart devices like smart phones,smart bands and watches which contain various sensors like accelerometer,gyroscope,GPS etc.,and adequate processing *** the task of recognition,human activities can be broadly categorized as basic and complex human *** of complex activities have received very less attention of researchers due to difficulty of problem by using either smart phones or smart *** reasons include lack of sensor-based labeled dataset having several complex human daily life *** of the researchers have worked on the smart phone’s inertial sensors to perform human activity recognition,whereas a few of them used both pocket and wrist *** this research,we have proposed a novel framework which is capable to recognize both basic and complex human activities using builtin-sensors of smart phone and smart *** have considered 25 physical activities,including 20 complex ones,using smart device’s built-in *** the best of our knowledge,the existing literature consider only up to 15 activities of daily life.

关键词： Complex human activities human daily life activities features extraction data fusion multi-sensory smartwatch smartphone

来源：评论

学校读者我要写书评

暂无评论

Parameter Estimation of PEM Fuel-Cell Using Brownian Butterfly Algorithm

Parameter Estimation of PEM Fuel-Cell Using Brownian Butterf...

引用

2024 International Conference on System, Computation, Automation and Networking, ICSCAN 2024

作者： Rajinikanth, Venkatesan Mohan, Ramya Narayanan, Mathiyazhagan Division of Research and Innovation Department of Computer Science and Engineering TN Chennai602105 India

ISBN: (纸本)9798331510022

The growing demand for electrical-energy (EE) has driven the development of varied power generation methods to meet global energy needs. The traditional Energy Generation Scheme (EGS) has greater impact on various environmental concerns and hence cleaner and sustainable energy sources are invented. Fuel-cell (FC) is one of the modern EGS, which converts the chemical-energy (CE) into EE with better efficiency. Further, the performance of the FC depends mainly on its parameters and hence it is necessary to identify the finest parameters. This research aims to develop a procedure to optimize the parameters of Nedstack-PS6;a commercially available Proton Exchange Membrane (PEM) FC (PEM-FC). The Nedstack-PS6 is a portable EE-source and its efficiency can be improved by assigning the finest possible value to its parameters. This research proposes a Brownian-Butterfly-Algorithm (BBA) based optimization process using a chosen cost-function for the chosen PEM-FC model. The BBA based search is implemented to solve a 7-dimensional optimization task using a bounded-search methodology and the achieved result with the BBA is compared against few chosen results by the optimization-algorithms. The experimental results confirms that the proposed scheme works well on the chosen PEM-FC model and help to identify the optimal values for all the 7-parametters. © 2024 IEEE.

关键词： Cost functions

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：