检索结果-内蒙古大学图书馆

Frankenstein Optimizer: Harnessing the Potential by Revisiting Optimization Tricks

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Hsu, Chia-Wei Tsou, Nien-Ti Chen, Yu-Cheng Park, Yang Jeong Li, Ju Department of Materials Science and Engineering Massachusetts Institute of Technology CambridgeMA02139 United States Department of Materials Science and Engineering National Yang Ming Chiao Tung University Taiwan Hon Hai Research Institute Taipei Taiwan Department of Nuclear Science and Engineering Massachusetts Institute of Technology CambridgeMA02139 United States

Gradient-based optimization drives the unprecedented performance of modern deep neural network models across diverse applications. adaptive algorithms have accelerated neural network training due to their rapid convergence rates;however, they struggle to find "flat minima" reliably, resulting in suboptimal generalization compared to stochastic gradient descent (SGD). By revisiting various adaptive algorithms’ mechanisms, we propose the Frankenstein optimizer, which combines their advantages. The proposed Frankenstein dynamically adjusts first- and second-momentum coefficients according to the optimizer’s current state to directly maintain consistent learning dynamics and immediately reflect sudden gradient changes. Extensive experiments across several research domains such as computer vision, natural language processing, few-shot learning, and scientific simulations show that Frankenstein surpasses existing adaptive algorithms and SGD empirically regarding convergence speed and generalization performance. Furthermore, this research deepens our understanding of adaptive algorithms through centered kernel alignment analysis and loss landscape visualization during the learning process. © 2025, CC BY.

关键词： adaptive algorithms

A Self-adaptive Harris Corner Detection Algorithm Based on Image contrast Areas

学校读者我要写书评

暂无评论

A Self-Adaptive Harris Corner Detection Algorithm Based on I...

Cyber-Energy Systems and Intelligent Energy (ICCSIE), International Conference on

作者： Yue Wang Meng-lei Xia Jun Pan Xiong Tang Nanjing Institute of Electronic Engineering Nanjing

ISBN: (数字)9798350373318

ISBN: (纸本)9798350373325

Largely for this reason that the threshold value of Harris algorithms can only be set by experience, corner is difficult to be contrasted in areas of image contrast. A new adaptive algorithm is proposed in this paper. Firstly, image pixels are selected by Sudoku method. Secondly, image contrast can be confirmed in area of above-mentioned pixels by calculating local mean square error. Finally, the best matching pixels according to image contrast are received by using self-adaption threshold, and experimental results can demonstrate the feasibility and effectiveness of the proposed adaptive Harris corner detection Algorithm.

关键词： adaptive systems Gaussian noise adaptive algorithms Mean square error methods Robustness Corner detection

Improved bounds for group testing in arbitrary hypergraphs

学校读者我要写书评

暂无评论

arXiv 2024年

作者： De Bonis, Annalisa Fisciano Italy

Recent papers initiated the study of a generalization of group testing where the potentially contaminated sets are the members of a given hypergraph F = (V, E). This generalization finds application in contexts where contaminations can be conditioned by some kinds of social and geographical clusterings. The paper focuses on few-stage group testing algorithms, i.e., slightly adaptive algorithms where tests are performed in stages and all tests performed in the same stage should be decided at the very beginning of the stage. In particular, the paper presents the first two-stage algorithm that uses o(d log |E|) tests for general hypergraphs with hyperedges of size at most d, and a three-stage algorithm that improves by a d1/6 factor on the number of tests of the best known three-stage algorithm. These algorithms are special cases of an s-stage algorithm designed for an arbitrary positive integer s ≤ d. The design of this algorithm resort to a new non-adaptive algorithm (one-stage algorithm), i.e., an algorithm where all tests must be decided beforehand. Further, we derive a lower bound for non-adaptive group testing. For E sufficiently large, the lower bound is very close to the upper bound on the number of tests of the best non-adaptive group testing algorithm known in the literature, and it is the first lower bound that improves on the information theoretic lower bound Ω(log |E|). © 2024, CC BY-NC-ND.

关键词： adaptive algorithms

Comparative Analysis of Autonomous Indoor Exploration Strategies: Floodfill algorithm vs. Frontier-Based Method

学校读者我要写书评

暂无评论

Comparative Analysis of Autonomous Indoor Exploration Strate...

International Conference on Control, Automation and Systems ( ICCAS)

作者： Naufaldo Hsiu-Ming Wu Graduate Institute of Mechatronic Engineering National Taipei University of Technology Taipei Taiwan Department of Intelligent Automation Engineering National Taipei University of Technology Taipei Taiwan

ISBN: (数字)9788993215380

ISBN: (纸本)9798331517939

This paper investigates the efficiency of autonomous indoor exploration utilizing simulation testing environments in Gazebo. Two exploration methods, Floodfill algorithm and Frontier-based algorithm, using the 2D LiDAR sensor are compared. The Floodfill algorithm employs a systematic traversal approach, while the Frontier-based method dynamically detects and navigates towards frontiers. Results indicate that the Frontier-based approach outperforms Flood-fill Algorithm in terms of efficiency and map completeness, particularly in complex environments. The study underscores the importance of the Frontier-based strategy for autonomous indoor exploration and paves the way for enhanced robotic applications in diverse domains.

关键词： Systematics Laser radar Automation Heuristic algorithms Indoor navigation adaptive algorithms Robot sensing systems Control systems Indoor environment Testing

A Modified Park Transformation Algorithm for Phase-Locked Loop Applications in Unbalanced Three-Phase Systems

学校读者我要写书评

暂无评论

A Modified Park Transformation Algorithm for Phase-Locked Lo...

IEEE Energy Conversion Congress and Exposition, ECCE

作者： Hsueh-Ju Wu Yaow-Ming Chen Department of Electrical Engineering National Taiwan University Taipei Taiwan

ISBN: (数字)9798350376067

ISBN: (纸本)9798350376074

This paper proposes a novel modified Park transformation (MPT) algorithm for phase-locked loop (PLL) applications in unbalanced three-phase systems. Unlike the conventional Park transformation, the proposed MPT algorithm directly incorporates unbalanced characteristics into its transformation matrix, thereby streamlining control strategies without requiring supplementary filtering. The effectiveness of the proposed MPT is verified through rigorous mathematical derivations, comparative simulations, and practical implementations using a digital signal processor (DSP). The results demonstrate that the MPT algorithm significantly improves the stability and accuracy of PLL systems under unbalanced conditions, providing an effective solution for handling phase unbalances in various power systems. Theoretical derivations and practical results, including crucial waveform comparisons, confirm the superior performance of the MPT algorithm and establish a robust framework for future research and applications.

关键词： Technological innovation Accuracy Heuristic algorithms Digital signal processors Signal processing algorithms adaptive algorithms Power system stability Stability analysis Phase locked loops Power harmonic filters

Tailoring Semantic Communication at Network Edge: A Novel Approach Using Dynamic Knowledge Distillation

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Albaseer, Abdullatif Abdallah, Mohamed Division of Information and Computing Technology College of Science and Engineering Hamad Bin Khalifa University Doha Qatar

Semantic Communication (SemCom) systems, empowered by deep learning (DL), represent a paradigm shift in data transmission. These systems prioritize the significance of content over sheer data volume. However, existing SemCom designs face challenges when applied to diverse computational capabilities and network conditions, particularly in time-sensitive applications. A key challenge is the assumption that diverse devices can uniformly benefit from a standard, large DL model in SemCom systems. This assumption becomes increasingly impractical, especially in high-speed, high-reliability applications such as industrial automation or critical healthcare. Therefore, this paper introduces a novel SemCom framework tailored for heterogeneous, resource-constrained edge devices and computation-intensive servers. Our approach employs dynamic knowledge distillation (KD) to customize semantic models for each device, balancing computational and communication constraints while ensuring Quality of Service (QoS). We formulate an optimization problem and develop an adaptive algorithm that iteratively refines semantic knowledge in edge devices, resulting in better models tailored to their resource profiles. This algorithm strategically adjusts the granularity of distilled knowledge, enabling devices to maintain high semantic accuracy for precise inference tasks, even under unstable network conditions. Extensive simulations demonstrate that our approach significantly reduces model complexity for edge devices, leading to better semantic extraction and achieving the desired QoS. © 2024, CC BY.

关键词： adaptive algorithms

An investigation on the use of Large Language Models for hyperparameter tuning in Evolutionary algorithms

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Custode, Leonardo Lucio Caraffini, Fabio Yaman, Anil Iacca, Giovanni University of Trento Trento Italy Swansea University Swansea United Kingdom Vrije Universiteit Amsterdam Amsterdam Netherlands

Hyperparameter optimization is a crucial problem in Evolutionary Computation. In fact, the values of the hyperparameters directly impact the trajectory taken by the optimization process, and their choice requires extensive reasoning by human operators. Although a variety of self-adaptive Evolutionary algorithms have been proposed in the literature, no definitive solution has been found. In this work, we perform a preliminary investigation to automate the reasoning process that leads to the choice of hyperparameter values. We employ two open-source Large Language Models (LLMs), namely Llama2-70b and Mixtral, to analyze the optimization logs online and provide novel real-time hyperparameter recommendations. We study our approach in the context of step-size adaptation for (1 + 1)-ES. The results suggest that LLMs can be an effective method for optimizing hyperparameters in Evolution Strategies, encouraging further research in this direction. Copyright © 2024, The Authors. All rights reserved.

关键词： adaptive algorithms

GOAL-ORIENTED ERROR ESTIMATION AND ADAPTIVITY FOR STOCHASTIC COLLOCATION FEM

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Bespalov, Alex Praetorius, Dirk Round, Thomas Savinov, Andrey School of Mathematics University of Birmingham Edgbaston BirminghamB15 2TT United Kingdom Institute of Analysis and Scientific Computing TU Wien Wiedner Hauptstraße 8–10 Vienna1040 Austria

We propose and analyze a general goal-oriented adaptive strategy for approximating quantities of interest (QoIs) associated with solutions to linear elliptic partial differential equations with random inputs. The QoIs are represented by bounded linear or continuously Gâteaux differentiable nonlinear goal functionals, and the approximations are computed using the sparse grid stochastic collocation finite element method (SC-FEM). The proposed adaptive strategy relies on novel reliable a posteriori estimates of the errors in approximating QoIs. One of the key features of our error estimation approach is the introduction of a correction term into the approximation of QoIs in order to compensate for the lack of (global) Galerkin orthogonality in the SC-FEM setting. Computational results generated using the proposed adaptive algorithm are presented in the paper for representative elliptic problems with affine and nonaffine parametric coefficient dependence and for a range of linear and nonlinear goal functionals. © 2024, CC BY.

关键词： adaptive algorithms

On an optimal AFEM for elastoplasticity

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Schönauer, Miriam Schröder, Andreas Paris Lodron University of Salzburg Department of Mathematics Salzburg Austria

In this paper, optimal convergence for an adaptive finite element algorithm for elastoplasticity is considered. To this end, the proposed adaptive algorithm is established within the abstract framework of the axioms of adaptivity [Comput. Math. Appl., 67(6) (2014), 1195-1253], which provides a specific proceeding to prove the optimal convergence of the scheme. The proceeding is based on verifying four axioms, which ensure the optimal convergence. The verification is done by using results from [Numer. Math., 132(1) (2016), 131-154], which presents an alternative approach to optimality without explicitly relying on the *** Codes 65N30 Copyright © 2024, The Authors. All rights reserved.

关键词： adaptive algorithms