检索结果-内蒙古大学图书馆

41st International Conference on Machine Learning, ICML 2024

作者： Li, Ziniu Xu, Tian Zhang, Yushun Lin, Zhihang Yu, Yang Sun, Ruoyu Luo, Zhi-Quan School of Data Science The Chinese University of Hong Kong Shenzhen China Shenzhen Research Institute of Big Data China National Key Laboratory for Novel Software Technology Nanjing University China Polixir.ai China China Shenzhen International Center For Industrial and Applied Mathematics China

Reinforcement Learning from Human Feedback (RLHF) is key to aligning Large Language Models (LLMs), typically paired with the Proximal Policy Optimization (PPO) algorithm. While PPO is a powerful method designed for general reinforcement learning tasks, it is overly sophisticated for LLMs, leading to laborious hyper-parameter tuning and significant computation burdens. To make RLHF efficient, we present ReMax, which leverages 3 properties of RLHF: fast simulation, deterministic transitions, and trajectory-level rewards. These properties are not exploited in PPO, making it less suitable for RLHF. Building on the renowned REINFORCE algorithm, ReMax does not require training an additional value model as in PPO and is further enhanced with a new variance reduction technique. ReMax offers several benefits over PPO: it is simpler to implement, eliminates more than 4 hyper-parameters in PPO, reduces GPU memory usage, and shortens training time. ReMax can save about 46% GPU memory than PPO when training a 7B model and enables training on A800-80GB GPUs without the memory-saving offloading technique needed by PPO. Applying ReMax to a Mistral-7B model resulted in a 94.78% win rate on the AlpacaEval leaderboard and a 7.739 score on MT-bench, setting a new SOTA for open-source 7B models. These results show the effectiveness of ReMax while addressing the limitations of PPO in LLMs. Copyright 2024 by the author(s)

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Dgi-Gru: Integrating Dynamic Graph Convolution for Enhanced Spatio-Temporal Information Extraction

SSRN

引用

SSRN 2024年

作者： Liu, Xinru Li, Weide Zhou, Da School of Mathematical Sciences National Institute for Data Science in Health and Medicine Xiamen University Xiamen China Laboratory of Applied Mathematics and Complex System Center of Applied Mathematics of Gansu Center for Data Science School of Mathematics and Statistics Lanzhou University Lanzhou China

The ubiquity of spatio-temporal data in the real world presents challenges for predictive modeling due to the intricate interplay of temporal trends and spatial correlations. In this paper, we address these challenges by proposing the DGI-GRU model, a novel approach that integrates graph convolution into a refined variant of the Gated Recurrent Unit (GRU) architecture, incorporating a dynamic adjacency matrix to capture evolving spatial relationships. The salient feature of the model is that it dynamically adjusts the adjacency matrix to account for time-dependent spatial correlations, enriching the representation of spatial information. Furthermore, our enhanced GRU framework enables simultaneous extraction of temporal and spatial features within a single-layer model, reducing parameter redundancy. Experimental results on real datasets demonstrate the superior performance of our proposed model in spatio-temporal sequence prediction tasks compared to state-of-the-art baselines. © 2024, The Authors. All rights reserved.

关键词： Convolution

来源：评论

学校读者我要写书评

暂无评论

Novel Scheme for Robust Confusion Component Selection Based on Pythagorean Fuzzy Set

引用

Computers, Materials & Continua 2023年第3期74卷 6523-6534页

作者： Nabilah Abughazalah Mohsin Iqbal Majid Khan Iqtadar Hussain Department of Mathematical Sciences College of SciencePrincess Nourah bint Abdulrahman UniversityP.O.Box 84428Riyadh11671Saudi Arabia Department of Mathematics&Statistics Riphah International UniversityIslamabadPakistan Department of Applied Mathematics&Statistics Institute of Space TechnologyIslamabadPakistan Mathematics Program Department of MathematicsStatistics and PhysicsCollege of Arts and SciencesQatar University2713DohaQatar Statistical Consulting Unit College of Arts and ScienceQatar UniversityDohaQatar

The substitution box,often known as an S-box,is a nonlinear component that is a part of several block *** purpose is to protect cryptographic algorithms from a variety of cryptanalytic assaults.A Multi-Criteria Decision Making(MCDM)problem has a complex selection procedure because of having many options and criteria to choose *** of this,statistical methods are necessary to assess the performance score of each S-box and decide which option is the best one available based on this *** the Pythagorean Fuzzy-based Technique for Order of Preference by Similarity to Ideal Solution(TOPSIS)method,the major objective of this investigation is to select the optimal S-box to be implemented from a pool of twelve key *** the help of the Pythagorean fuzzy set(PFS),the purpose of this article is to evaluate whether this nonlinear component is suitable for use in a variety of encryption *** this article,we have considered various characteristics of S-boxes,including nonlinearity,algebraic degree,strict avalanche criterion(SAC),absolute indicator,bit independent criterion(BIC),sum of square indicator,algebraic immunity,transparency order,robustness to differential cryptanalysis,composite algebraic immunity,signal to noise ratio-differential power attack(SNR-DPA),and confusion coefficient variance on some standard S-boxes that are Advanced Encryption Following this,the findings of the investigation are changed into Pythagorean fuzzy numbers in the shape of a *** matrix is then subjected to an analysis using the TOPSIS method,which is dependent on the Pythagorean fuzzy set,to rank the most suitable S-box for use in encryption applications.

关键词： Decision making substitution box TOPSIS multi-criterion decision making fuzzy set Pythagorean fuzzy set

来源：评论

学校读者我要写书评

暂无评论

Bio-heat transfer in skin tissue using fractional Weinbaum-Jiji model

Bio-heat transfer in skin tissue using fractional Weinbaum-J...

引用

2023 International Conference on Fractional Differentiation and Its Applications, ICFDA 2023

作者： Essam, R. Elsaid, A. Zahra, W.K. Egypt-Japan University of Science and Technology Institute of Basic and Applied Sciences Alexandria Egypt Mansoura University Faculty of Engineering Mathematics and Engineering Physics Department Mansoura Egypt Tanta University Faculty of Engineering Engineering Physics and Mathematics Department Tanta Egypt

ISBN: (纸本)9798350321685

This study uses the Weinbaum-Jiji bioheat model to numerically simulate heat transfer through skin tissue that is exposed to a constant heat flux. The classical Weinbaum-Jiji equation is modified by including the Caputo fractional derivative in order to simulate the finite thermal propagation speed. The integration of the fractional calculus into the classical Weinbaum-Jiji equation facilitates the characterization of the flux-gradient relationship in terms of thermal time lag. This advancement extends the capabilities of the equation, allowing for a deeper understanding of the underlying physical processes. Using a discrete scheme, the problem is solved using a finite difference method for time and a finite element method for space. The results are compared to the previous efforts to model the problem, and good agreement is found in the rest of the case. The effects of the in-flow Péclet number and the effective thermal conductivity on the thermal profile along the tissues are illustrated. The variation of thermal profile along with the change in the inflow Péclet number reveals how vital is taking the vascular parameters into consideration. © 2023 IEEE.

关键词： Finite difference method

来源：评论

学校读者我要写书评

暂无评论

HIGH-DIMENSIONAL SGD ALIGNS WITH EMERGING OUTLIER EIGENSPACES 12

HIGH-DIMENSIONAL SGD ALIGNS WITH EMERGING OUTLIER EIGENSPACE...

引用

12th International Conference on Learning Representations, ICLR 2024

作者： Arous, Gérard Ben Gheissari, Reza Huang, Jiaoyang Jagannath, Aukosh Courant Institute New York University New YorkNY United States Department of Mathematics Northwestern University EvanstonIL United States Department of Statistics and Data Science University of Pennsylvania PhiladelphiaPA United States Department of Statistics and Actuarial Science Department of Applied Mathematics Cheriton School of Computer Science University of Waterloo WaterlooON Canada

We rigorously study the joint evolution of training dynamics via stochastic gradient descent (SGD) and the spectra of empirical Hessian and gradient matrices. We prove that in two canonical classification tasks for multi-class high-dimensional mixtures and either 1 or 2-layer neural networks, the SGD trajectory rapidly aligns with emerging low-rank outlier eigenspaces of the Hessian and gradient matrices. Moreover, in multi-layer settings this alignment occurs per layer, with the final layer's outlier eigenspace evolving over the course of training, and exhibiting rank deficiency when the SGD converges to sub-optimal classifiers. This establishes some of the rich predictions that have arisen from extensive numerical studies in the last decade about the spectra of Hessian and information matrices over the course of training in overparametrized networks. © 2024 12th International Conference on Learning Representations, ICLR 2024. All rights reserved.

关键词： Statistics

来源：评论

学校读者我要写书评

暂无评论

Electron stopping power of protons in aluminum at finite electron temperature

引用

Physical Review B 2025年第14期111卷 144303-144303页

作者： Ya-Ting Sun Feng Wang Fei Mao Cong-Zhang Gao School of Physics Beijing Institute of Technology Beijing 100081 China School of Nuclear Science and Technology University of South China Hengyang 421001 China Institute of Applied Physics and Computational Mathematics Beijing 100088 China

We conducted a study on the electron stopping power of protons in aluminum at finite electron temperatures, utilizing time-dependent density functional theory nonadiabatically coupled with molecular dynamics. Our investigation focused on protons with initial velocities ranging from 0.1 to 1.0 a.u., providing a wealth of detailed information on the electronic states involved in the stopping process, with exceptional spatial and temporal resolution. Our results show that the electron temperature can significantly effect the electron stopping power. A quantum-blocking mechanism based on a physical picture of electronic transitions in energy levels has been proposed for explaining the phenomenon of electron stopping power decreasing with the increase of target electron temperature.

关键词： Electronic states

来源：评论

学校读者我要写书评

暂无评论

Analysis of Import Substitution Processes Taking into Account Industry Specifics in the Regional Economy 14th

Analysis of Import Substitution Processes Taking into Acco...

引用

14th International Conference on Optimization and Applications, OPTIMA 2023

作者： Rassokha, Anastasiia Obrosova, Natalia Moscow Institute of Physics and Technology Dolgoprudny Russia Federal Research Center "Computer Science and Control" of RAS Moscow Russia Center of Fundamental and Applied Mathematics Lomonosov MSU Moscow Russia

ISBN: (纸本)9783031478581

Analysis of import substitution processes is one of the urgent problems for many countries in the context of deglobalization of the world economy. Mathematical methods of convex analysis make it possible to construct a model of input-output balance with non-linear production functions. It can be identified from the official statistics of input-output tables. This model allows us to describe the substitution of production factors. The elasticity of substitution can be estimated using the solution of the Yang dual problem. With the help of the developed model for industries with a significant share of imports, the possibility of replacing imported production factors with domestic ones is studied on the example of the Russian economy. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Regional planning

来源：评论

学校读者我要写书评

暂无评论

Optimized Vaccine Selection Using Machine Learning and Genetic Algorithms: A Study on Side Effects of COVID-19 Vaccines 12th

Optimized Vaccine Selection Using Machine Learning and Genet...

引用

12th International Conference on Soft Computing for Problem Solving, SocProS 2023

作者： Soni, Vishal Joshi, Shubham Deep, Kusum Pant, Millie Department of Mechatronics Indian Institute of Information Technology Bihar Bhagalpur813210 India Department of Applied Mathematics and Scientific Computing Indian Institute of Technology Uttarakhand Roorkee247667 India Department of Mathematics Indian Institute of Technology Uttarakhand Roorkee247667 India Mehta Family School for Data Science and Artificial Intelligence Indian Institute of Technology Uttarakhand Roorkee247667 India

ISBN: (纸本)9789819732913

The COVID-19 pandemic has necessitated large-scale vaccination campaigns to control the virus's spread. Given the scale of these efforts, it's crucial to scrutinize potential adverse effects, especially in vulnerable population groups. This study combines machine learning and Genetic Algorithms (GA) to analyze the patterns of patient recovery post-vaccination. We use a range of classifiers, including the Gradient Boosting Decision Tree (GBDT), to identify key features and symptom categories that influence patient recovery. We also use Principal Component Analysis (PCA) to reduce dimensionality and uncover patterns within high-dimensional data. The data used in this study have taken from the Vaccine Adverse Event Reporting System (VAERS). With the help of GA, we aim to guide optimal vaccine selection for susceptible demographics, minimizing the risk and severity of side effects. Our research paves the way for more personalized vaccination strategies, promoting healthier outcomes and reducing risks associated with mass vaccination. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.

关键词： COVID-19

来源：评论

学校读者我要写书评

暂无评论

DCE MRI Modality Investigation for Cancerous Prostate Region Detection: Case Analysis 30

DCE MRI Modality Investigation for Cancerous Prostate Region...

引用

30th International Conference in Central Europe on Computer Graphics, Visualization and Computer Vision, WSCG 2022

作者： Vaitulevicius, Aleksas Treigys, Povilas Bernataviciene, Jolita Surkant, Roman Markeviciute, Jurgita Naruseviciute, Ieva Trakymas, Mantas Vilnius University Institute of Data Science and Digital Technologies Vilnius Lithuania Vilnius University Institute of Applied Mathematics Vilnius Lithuania National Cancer Institute Vilnius Lithuania

ISBN: (纸本)9788086943336

Typically, prostate evaluation is done by using different imaging sequences of magnetic resonance imaging. Dynamic contrast enhancement, one of such scanning modalities, allow to spot higher vascular permeability and density caused by the malignant tissue. Authors of this paper investigate the ability to identify malignant prostate regions by the functional data analysis and standard machine learning techniques. The dynamic contrast enhanced images of the prostate are divided into the regions and based on those time-signal intensity curves are calculated. Two classification approaches: functional k-Nearest Neighbors and machine learning Support Vector Machine are used to model signal curve behavior on temporal variation matrix and timestamp based prostate region division of image data. Preliminary research shows that both functional data analysis and machine learning classification methods are able to identify highest saturation timestamp that gives best tissue classification results on timestamp based dynamic contrast enhanced region map obtained by Simple Linear Iterative Clustering algorithm. Cancer region classification results are better when the dynamic contrast enhanced images are subdivided into regions at each timestamp than when using a temporal variation matrix. © 2022 University of West Bohemia. All rights reserved.

关键词： Dynamic contrast enhanced MRI

来源：评论

学校读者我要写书评

暂无评论

Entropies of the Y-Junction Type Nanostructures

引用

Computer Modeling in Engineering & sciences 2023年第6期135卷 2665-2679页

作者： Ricai Luo Aisha Javed Muhammad Azeem Muhammad Kamran Jamil Hassan Raza Muhammad Yasir Ilyas School of Mathematics and Physics Hechi UniversityYizhou456300China Abdus Salam School of Mathematical Sciences Government College UniversityLahore54000Pakistan Department of Mathematics Riphah Institute of Computing and Applied SciencesRiphah International UniversityLahore54000Pakistan School of Mathematical Sciences College of Science and TechnologyWenzhou-Kean UniversityWenzhou325060China Department of Physics COMSATS University IslamabadLahore54000Pakistan

Recent research on nanostructures has demonstrated their importance and application in a variety of *** are used directly or indirectly in drug delivery systems,medicine and pharmaceuticals,biological sensors,photodetectors,transistors,optical and electronic devices,and so *** discovery of carbon nanotubes with Y-shaped junctions is motivated by the development of future advanced electronic *** of their interactionwithY-junctions,electronic switches,amplifiers,and three-terminal transistors are of particular *** is a concept that determines the uncertainty of a system or *** concepts are also used in biology,chemistry,and applied *** on the requirements,entropy in the form of a graph can be classified into several *** 1955,graph-based entropy was *** of the types of entropy is edgeweighted *** examined the abstract form of Y-shaped junctions in this *** edge-weight-based entropy formulas for the generic view of Y-shaped junctions were created,and some edge-weighted and topological index-based concepts for Y-shaped junctions were discussed in the present paper.

关键词： Y-shaped junctions three-terminal junctions nanostuctures entropy carbon nanotube

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：