检索结果-内蒙古大学图书馆

arXiv 2025年

作者： Sabolić, Ivan Grcić, Matej Šegvić, Siniša University of Zagreb Faculty of Electrical Engineering and Computing Unska 3 Zagreb10000 Croatia

We propose VIBE, a model-agnostic framework that trains classifiers resilient to backdoor attacks. The key concept behind our approach is to treat malicious inputs and corrupted labels from the training dataset as observed random variables, while the actual clean labels are latent. VIBE then recovers the corresponding latent clean label posterior through variational inference. The resulting training procedure follows the expectation-maximization (EM) algorithm. The E-step infers the clean pseudolabels by solving an entropy-regularized optimal transport problem, while the M-step updates the classifier parameters via gradient descent. Being modular, VIBE can seamlessly integrate with recent advancements in self-supervised representation learning, which enhance its ability to resist backdoor attacks. We experimentally validate the method effectiveness against contemporary backdoor attacks on standard datasets, a large-scale setup with 1k classes, and a dataset poisoned with multiple attacks. VIBE consistently outperforms previous defenses across all tested scenarios. © 2025, CC BY-NC-SA.

关键词： Expectation maximization algorithm

来源：评论

学校读者我要写书评

暂无评论

Large Language Models as Attribution Regularizers for Efficient Model Training

arXiv

引用

arXiv 2025年

作者： Vukadin, Davor Šilić, Marin Delač, Goran University of Zagreb Faculty of Electrical Engineering and Computing 3 Unska street Zagreb Croatia

Large Language Models (LLMs) have demonstrated remarkable performance across diverse domains. However, effectively leveraging their vast knowledge for training smaller downstream models remains an open challenge, especially in domains like tabular data learning, where simpler models are often preferred due to interpretability and efficiency. In this paper, we introduce a novel yet straightforward method for incorporating LLM-generated global task feature attributions into the training process of smaller networks. Specifically, we propose an attribution-matching regularization term that aligns the training dynamics of the smaller model with the insights provided by the LLM. By doing so, our approach yields superior performance in few-shot learning scenarios. Notably, our method requires only black-box API access to the LLM, making it easy to integrate into existing training pipelines with minimal computational overhead. Furthermore, we demonstrate how this method can be used to address common issues in real-world datasets, such as skewness and bias. By integrating high-level knowledge from LLMs, our approach improves generalization, even when training data is limited or imbalanced. We validate its effectiveness through extensive experiments across multiple tasks, demonstrating improved learning efficiency and model robustness. © 2025, CC BY-NC-SA.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Propagation of Electromagnetic Waves Along Cylindrical Metasurface Structure

Propagation of Electromagnetic Waves Along Cylindrical Metas...

引用

European Conference on Antennas and Propagation, EuCAP

作者： Davorin Mikulić Marko Bosiljevac Christophe Craeye Zvonimir Šipuš Faculty of Electrical Engineering and Computing University of Zagreb Zagreb Croatia UCLouvain Louvain-la-Neuve Belgium

ISBN: (数字)9788831299107

ISBN: (纸本)9798350366327

This paper explores the design of cylindrical leaky-wave metasurface antennas, addressing the need for antennas that conform to curved cylindrical surfaces. We introduce analytical tools for efficiently extracting the properties of electromagnetic waves propagating along cylindrical metasurfaces, enabling the suppression of curvature effects in the antenna design. A comprehensive design procedure is presented, along with a detailed design example and its corresponding simulation results. Our approach provides a fast and efficient method for designing cylindrical leaky-wave metasurface antennas, enabling the ability to radiate in the desired direction.

关键词： Surface impedance Surface waves Simulation Electromagnetic scattering Europe Directive antennas Metasurfaces Leaky wave antennas Impedance Electromagnetics

来源：评论

学校读者我要写书评

暂无评论

Quality-diversity and Novelty Search for Portfolio Optimization and Beyond

引用

Computational Economics 2025年 1-24页

作者： Gašperov, Bruno Begušić, Stjepan Bauman, Tessa Kostanjčar, Zvonko Faculty of Computer and Information Science University of Ljubljana Večna pot 113 Ljubljana Slovenia Faculty of Electrical Engineering and Computing University of Zagreb Unska 3 Zagreb Croatia

In this paper, we provide an overview of exploration algorithms, a novel class of evolutionary computation approaches encompassing novelty search and quality-diversity, as they may be applied to portfolio optimization and related problems in quantitative finance. These algorithms rely on the idea of divergent search, prioritizing the discovery of novel, unexplored solutions over maximizing a classical objective (fitness) function. Their favorable exploration properties make them particularly valuable in evolutionary reinforcement learning, where they have been successfully applied to diverse practical domains such as procedural generation of video game content and robotics. However, their application in portfolio optimization (PO) and quantitative finance in general, remains largely unexplored. We posit that this gap presents significant untapped potential, given the natural connections between divergent search principles and key concepts in PO, including diversification, inter-individual variety of risk preferences, and sensitivity to market regimes. Following an introduction to exploration algorithms, we delve into existing research on their applications in PO, outlining their advantages and shortcomings. Moreover, we propose several research directions that are expected to pave the way for more substantial leverage of such algorithms in decision-making processes in PO and beyond, while emphasizing challenges that stand in the way of their realization. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025.

关键词： Evolutionary computation Exploration algorithms Novelty search Portfolio optimization Quality-diversity Quantitative finance

来源：评论

学校读者我要写书评

暂无评论

Graph-Based Decentralized Exploration and Semantic Inspection for Multi-Robot Systems

Graph-Based Decentralized Exploration and Semantic Inspectio...

引用

International Conference on Unmanned Aircraft Systems (ICUAS)

作者： Nada E. Abbas Tamara Petrovic Faculty of Electrical Engineering and Computing LARICS Laboratory for Robotics and Intelligent Control Systems University of Zagreb Zagreb Croatia

ISBN: (数字)9798331513283

ISBN: (纸本)9798331513290

This paper presents a decentralized graph-based exploration and inspection framework for Multi-Robot Systems, designed to address challenges in subterranean and largescale environments. Unlike prior works that focus solely on exploration or inspection, this framework integrates volumetric exploration, semantic inspection, and dynamic task allocation into a unified decentralized system. A key novelty of this work is the seamless integration of these modules in a multi-robot setting, allowing UAVs to autonomously coordinate their tasks without relying on centralized control. The framework employs a hierarchical graph structure, utilizing a dense local graph for immediate navigation and a sparse global graph for long-term planning and repositioning. Extensive simulations in large-scale complex-shaped environments demonstrate that the proposed approach improves the completeness of the generated maps, reduces inconsistencies in the constructed mesh, and accelerates the overall exploration-inspection process compared to existing decentralized strategies.

关键词： Semantics Inspection Prediction algorithms Dynamic scheduling Aircraft navigation Robustness Planning Multi-robot systems Resource management Centralized control

来源：评论

学校读者我要写书评

暂无评论

Genetic programming with surrogate evaluation for the electric vehicle routing problem

引用

Swarm and Evolutionary Computation 2025年 96卷

作者： Gil-Gala, Francisco J. Đurasević, Marko Jakobović, Domagoj Varela, Ramiro Department of Computer Science University of Oviedo Gijón33271 Spain Faculty of Electrical Engineering and Computing University of Zagreb Zagreb10000 Croatia

The focus on environmental sustainability has made the Electric Vehicle Routing Problem (EVRP) an important area of research. Routing Policies (RPs) offer a simple and efficient approach to solving VRPs, providing advantages over methods like metaheuristics by quickly generating solutions. However, designing efficient RPs manually can be time-consuming. Therefore, there is a need to explore hyper-heuristic approaches, particularly Genetic Programming (GP), to automate the design of RPs. However, population-based evolutionary algorithms like GP often require a significant amount of computational resources, especially for fitness calculation. Therefore, surrogate evaluation is essential in enhancing efficiency, especially in GP, where multiple problem instances need to be solved to evaluate each chromosome. In this study, we employ surrogate models within GP to design RPs for EVRP with hard time windows. The experiments show that the RPs designed by GP with surrogate models outperform those produced by standard GP approaches while still requiring less computational time to be generated. Moreover, the RPs designed with GP using surrogate models are also smaller, and consequently, they are also more efficient and easier to interpret. © 2025

关键词： Genetic programming

来源：评论

学校读者我要写书评

暂无评论

Designing Lookahead Relocation Rules for the Container Relocation Problem with Genetic Programming 28th

Designing Lookahead Relocation Rules for the Container Rel...

引用

28th European Conference on Genetic Programming, EuroGP 2025, Held as Part of EvoStar 2025

作者： Đurasević, Marko Đumić, Mateja Gil-Gala, Francisco Javier Jakobović, Domagoj Faculty of Electrical Engineering and Computing University of Zagreb Zagreb10000 Croatia School of Applied Mathematics and Informatics J. J. Strossmayer University of Osijek Osijek31000 Croatia Department of Computing University of Oviedo Gijón Spain

ISBN: (纸本)9783031899904

The container relocation problem is an important combinatorial optimisation problem commonly found in warehouses and container ports. The goal of this problem is to retrieve all of the containers from the yard with the fewest container relocations between the stacks. Since the problem is NP-hard, various heuristics have been proposed to solve it, among which are relocation rules (RRs), simple constructive heuristics that incrementally construct the solution. However, it is quite difficult to design such RRs manually, so genetic programming has been applied to design new RRs automatically. A significant problem with RRs, whether manually or automatically designed, is that they usually have a limited view of the problem. This means that they will often make decisions that can negatively influence the future, meaning that the current decision would cause additional relocations. Therefore, this study investigates different relocation schemes that can be used within RRs to obtain rules with lookahead ability. These rules will enable containers to be relocated based on future information and, consequently, arranged better in the yard. For that purpose, three novel relocation schemes for automatically designed RRs are defined and evaluated on an existing problem set. The results demonstrate that integrating additional elements to evolve lookahead RRs can significantly improve the results. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Genetic programming

来源：评论

学校读者我要写书评

暂无评论

Reactive Navigation of the Ackermann Steering Robot in Unknown Environments

Reactive Navigation of the Ackermann Steering Robot in Unkno...

引用

International Conference on Autonomous Robots and Agents, ICARA

作者： ANA ŠELEK IVAN HRABAR SAŠA ŠTEKOVIĆ Department of Control and Computer Engineering Faculty of Electrical Engineering and Computing University of Zagreb Zagreb Croatia Autegra D.O.O. Zagreb Croatia

ISBN: (数字)9798331509231

ISBN: (纸本)9798331509248

Mobile robots are now widely used in numerous real-world applications that have complex navigation requirements, especially in environments used by humans. This requires highly accurate navigation that can be performed in realtime. In this paper, a method for generating a smooth motion of nonholonomic mobile robots is proposed. It enables robots to move optimally toward the desired goal and allows fast path replanning when encountering unknown or dynamic obstacles. The method generates smooth, collision-free trajectories based on clothoids, ensuring high computational efficiency and suitability for realtime path planning. By applying a smoothing algorithm, the proposed method improves the robot's efficiency in terms of travel time and trajectory length from start to goal, as demonstrated by a comparison with model predictive control.

关键词： Smoothing methods Navigation Dynamics Real-time systems Trajectory Computational efficiency Mobile robots Collision avoidance Testing Predictive control

来源：评论

学校读者我要写书评

暂无评论

Power Line Following Based on Measurements of the Magnetic Field

Power Line Following Based on Measurements of the Magnetic F...

引用

International Conference on Unmanned Aircraft Systems (ICUAS)

作者： Goran Vasiljević Dean Martinović Stjepan Bogdan Faculty of Electrical Engineering and Computing Laboratory for Robotics and Intelligent Control Systems Unska 3 University of Zagreb Zagreb Croatia

ISBN: (数字)9798331513283

ISBN: (纸本)9798331513290

In this paper, we present a method to control the UAV to follow the power line with a specific position and orientation based only on the measurement of the magnetic field generated by the current flow in the power line. In this way, it is possible to localize the UAV with respect to the power line without the need for additional sensors, even in poor visibility conditions. The measurements from four magnetometers attached to the UAV are used to solve an optimization problem that involves determining the relative pose of the UAV with respect to the power line. Based on the relative pose, the UAV is controlled to follow the power line in a predefined position and orientation. Experiments in a test setup have confirmed that the method is applicable in a realistic environment.

关键词： Location awareness Visualization Power measurement Magnetic field measurement Magnetometers Current measurement Position measurement Autonomous aerial vehicles Magnetic fields Optimization

来源：评论

学校读者我要写书评

暂无评论

Pose-independent 3D Anthropometry from Sparse Data

arXiv

引用

arXiv 2025年

作者： Bojanić, David Wuhrer, Stefanie Petković, Tomislav Pribanić, Tomislav University of Zagreb Faculty of Electrical Engineering and Computing Zagreb Croatia Inria centre University Grenoble Alpes Grenoble France

3D digital anthropometry is the study of estimating human body measurements from 3D scans. Precise body measurements are important health indicators in the medical industry, and guiding factors in the fashion, ergonomic and entertainment industries. The measuring protocol consists of scanning the whole subject in the static A-pose, which is maintained without breathing or movement during the scanning process. However, the A-pose is not easy to maintain during the whole scanning process, which can last even up to a couple of minutes. This constraint affects the final quality of the scan, which in turn affects the accuracy of the estimated body measurements obtained from methods that rely on dense geometric data. Additionally, this constraint makes it impossible to develop a digital anthropometry method for subjects unable to assume the A-pose, such as those with injuries or disabilities. We propose a method that can obtain body measurements from sparse landmarks acquired in any pose. We make use of the sparse landmarks of the posed subject to create pose-independent features, and train a network to predict the body measurements as taken from the standard A-pose. We show that our method achieves comparable results to competing methods that use dense geometry in the standard A-pose, but has the capability of estimating the body measurements from any pose using sparse landmarks only. Finally, we address the lack of open-source 3D anthropometry methods by making our method available to the research community at ***/DavidBoja/pose-independent-anthropometry. Copyright © 2025, The Authors. All rights reserved.

关键词： Entertainment industry

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：