检索结果-内蒙古大学图书馆

Secure and Trustworthy Machine Learning (SaTML), IEEE Conference on

作者： Daryna Oliynyk Rudolf Mayer Andreas Rauber CDL AsTra Faculty of Computer Science University of Vienna Vienna Austria SBA Research Vienna Austria Faculty of Informatics TU Wien Institute of Information Systems Engineering Vienna Austria

ISBN: (数字)9798331517113

ISBN: (纸本)9798331517120

Machine learning (ML) models were shown to be vulnerable to model stealing attacks, which lead to intellectual property infringement. Among other attack methods, substitute model training is an all-encompassing attack applicable to any machine learning model whose behaviour can be approximated from input-output queries. Whereas previous works mainly focused on improving the performance of substitute models by, e.g. developing a new substitute training method, there have been only limited ablation studies that try to understand the impact the strength of an attacker has on the substitute model's performance. As a result, different authors came to diverse, sometimes contradicting, conclusions. In this work, we exhaustively examine the ambivalent influence of different factors resulting from varying the attacker's capabilities and knowledge on a substitute training attack. Our findings suggest that some of the factors that have been considered important in the past are, in fact, not that influential; instead, we discover new correlations between attack conditions and success rate. In particular, we demonstrate that better-performing target models enable higher-fidelity attacks and explain the intuition behind this phenomenon. Further, we propose to shift the focus from the complexity of target models toward the complexity of their learning tasks. Therefore, for the substitute model, rather than aiming for a higher architecture complexity, we suggest focusing on getting data of higher complexity and an appropriate architecture. Finally, we demonstrate that even in the most limited data-free scenario, there is no need to overcompensate weak knowledge with unrealistic capabilities in the form of millions of queries. Our results often exceed or match the performance of previous attacks that assume a stronger attacker, suggesting that these stronger attacks are likely endangering a model owner's intellectual property to a significantly higher degree than shown until now.

关键词： Training Correlation Query processing Focusing Intellectual property Data collection Data models Adversarial machine learning Complexity theory Tuning

来源：评论

学校读者我要写书评

暂无评论

SEMU: Singular Value Decomposition for Efficient Machine Unlearning

arXiv

引用

arXiv 2025年

作者： Sendera, Marcin Struski, Lukasz Ksiazek, Kamil Musiol, Kryspin Tabor, Jacek Rymarczyk, Dawid Faculty of Mathematics and Computer Science Jagiellonian University Poland Institute of Theoretical and Applied Informatics Polish Academy of Sciences Poland Ardigen SA Poland

While the capabilities of generative foundational models have advanced rapidly in recent years, methods to prevent harmful and unsafe behaviors remain underdeveloped. Among the pressing challenges in AI safety, machine unlearning (MU) has become increasingly critical to meet upcoming safety regulations. Most existing MU approaches focus on altering the most significant parameters of the model. However, these methods often require fine-tuning substantial portions of the model, resulting in high computational costs and training instabilities, which are typically mitigated by access to the original training dataset. In this work, we address these limitations by leveraging Singular Value Decomposition (SVD) to create a compact, low-dimensional projection that enables the selective forgetting of specific data points. We propose Singular Value Decomposition for Efficient Machine Unlearning (SEMU), a novel approach designed to optimize MU in two key aspects. First, SEMU minimizes the number of model parameters that need to be modified, effectively removing unwanted knowledge while making only minimal changes to the model’s weights. Second, SEMU eliminates the dependency on the original training dataset, preserving the model’s previously acquired knowledge without additional data requirements. © 2025, CC BY.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A Hybrid Adaptive Filter for Head Tracking in Augmented Reality (AR)-based Flight Simulators

引用

IEEE Transactions on computers 2025年

作者： Nwobodo, Onyeka J. Kuaban, Godlove Suila Nkemeni, Valery Wereszczynski, Kamil Cyran, Krzysztof A. Silesian University of Technology Department of Computer Graphics Vision and Digital Systems Faculty of Automatic Control Electronics and Computer Science Gliwice Poland Polish Academy of Sciences Institute of Theoretical and Applied Informatics Poland University of Buea Laboratory of Electrical Engineering and Computing Faculty of Engineering and Technology Buea Cameroon

This paper presents a novel approach for head tracking in augmented reality (AR) flight simulators using an adaptive fusion of Kalman and particle filters. This fusion dynamically balances the strengths of both algorithms, leveraging Kalman filters for computational efficiency and particle filters for handling non-linearities based on real-time factors such as sensor noise and head movement patterns. Our method demonstrates superior tracking precision and reduced latency, making it particularly effective for immersive pilot training in AR-based flight simulations. While focused on flight simulation, our approach holds high potential for broader applications in other AR and virtual reality (VR) environments where precise, real-time head tracking is crucial. These results provide actionable design guidance for developers optimizing tracking systems in environments that require fast response times and high accuracy. Future extensions of this research could explore the generalization of our approach across diverse AR applications, including human-computer interaction (HCI), medical simulations, and gaming. © 1968-2012 IEEE.

关键词： Flight simulators

来源：评论

学校读者我要写书评

暂无评论

The χ-Binding Function of d-Directional Segment Graphs

引用

Discrete and Computational Geometry 2025年 1-13页

作者： Duraj, Lech Kang, Ross J. La, Hoang Narboni, Jonathan Pokrývka, Filip Rambaud, Clément Reinald, Amadeus Theoretical Computer Science Department Faculty of Mathematics and Computer Science Kraków Poland Korteweg-de Vries Institute for Mathematics University of Amsterdam Amsterdam Netherlands Faculty of Informatics Masaryk University Brno Czech Republic Université Côte d’Azur CNRS Inria I3S Sophia-Antipolis France

Given a positive integer d, the class d-DIR is defined as all those intersection graphs formed from a finite collection of line segments in R2 having at most d slopes. Since each slope induces an interval graph, it easily follows for every G in d-DIR with clique number at most ω that the chromatic number χ(G) of G is at most dω. We show for every even value of ω how to construct a graph in d-DIR that meets this bound exactly. This partially confirms a conjecture of Bhattacharya, Dvořák and Noorizadeh. Furthermore, we show that the χ-binding function of d-DIR is ω↦dω for ω even and ω↦d(ω-1)+1 for ω odd. This extends an earlier result by Kostochka and Nešetřil, which treated the special case d=2. © The Author(s) 2025.

关键词： Chi-boundedness Geometric graphs Graph colouring Segment graphs

来源：评论

学校读者我要写书评

暂无评论

Face image authentication scheme based on MTCNN and SLT

引用

Multimedia Tools and Applications 2025年 1-43页

作者： Thabit, Rasha Al-Askari, Mohanad A. Mohammed, Dunya Zeki Anaam, Elham Abdulwahab Mahmood, Zainab H. Jabbar, Dina Jamal Salih, Zahraa Aqeel Information System Department College of Computer Sciences and Information Technology University of Al-Anbar Al-Anbar Iraq Computer Engineering Department College of Engineering Al-Iraqia University Baghdad Iraq Electronic and Communications Engineering Department College of Engineering Gilgamesh University Baghdad Iraq Faculty of Computing and Informatics Multimedia University Selangor Cyberjaya Malaysia Department of Cybersecurity Engineering College of Information Engineering Al-Nahrain University Baghdad Iraq Electrical Engineering Technical Collage Middle Technical University Baghdad Iraq Department of Business Administration Alamaal University College Baghdad Iraq

DeepFakes and face image manipulation methods have been widely distributed in the last few years and several techniques have been presented to check the authenticity of the face image and detect manipulation if exists. Most of the available manipulation detection techniques have been successfully applied to reveal one type of manipulation under specific conditions, however, many limitations and challenges can be encountered in this field. To overcome some limitations and challenges, this paper presents a new face image authentication (FIA) scheme based on Multi-Task Cascaded Conventional Neural Networks (MTCNN) and watermarking in Slantlet transform (SLT) domain. The proposed FIA scheme has three main algorithms that are face detection and selection, embedding, and extraction algorithms. Different block sizes have been used to divide the image into non-overlapping blocks followed by classifying them into two groups that are blocks from face area (FA) and blocks from the remaining area (RA) of the image. In the embedding algorithms, the authentication information is generated from FA blocks and embedded in the RA blocks. In the extraction algorithms, the embedded information is extracted from RA blocks and compared with the calculated data from FA blocks to reveal manipulations and localize the manipulated blocks if exist. Extensive experiments have been conducted to evaluate the performance of the proposed FIA scheme for different face images. The experimental work included tests for payload, capacity, visual quality, time complexity, and localization of manipulations. The results proved the efficiency of the proposed scheme in detecting and localizing different face image manipulations such as attributes attacks, retouching attacks, expression swap, face swap, and morphing attacks. The proposed scheme overcomes many limitations and it is 100% accurate in localizing the tampered blocks which makes it a better candidate for practical applications. © The Author(s), un

关键词： Authentication

来源：评论

学校读者我要写书评

暂无评论

Generative AI in Academic Writing: A Comparison of DeepSeek, Qwen, ChatGPT, Gemini, Llama, Mistral, and Gemma

arXiv

引用

arXiv 2025年

作者： Aydın, Ömer Karaarslan, Enis Erenay, Fatih Safa Džakula, Nebojša Bačanin Manisa Celal Bayar University Faculty of Engineering and Natural Sciences Department of Computer Sciences Manisa Turkey University of Waterloo Faculty of Engineering Management Science and Engineering Ontario Canada Mugla Sitki Kocman University Faculty of Engineering Department of Computer Engineering Mugla Turkey Singidunum University Faculty of Informatics and Computing Belgrade Serbia

Deepseek and Qwen LLMs became popular at the beginning of 2025 with their low-cost and open-access LLM solutions. A company based in Hangzhou, Zhejiang, China, announced its new LLM, DeepSeek v3, in December 2024. Then, Alibaba released its AI model, Qwen 2.5 Max, on January 29, 2025. These tools, which are free and open-source have made a significant impact on the world. Deepseek and Qwen also have the potential to be used by many researchers and individuals around the world in academic writing and content creation. Therefore, it is important to determine the capacity of these new LLMs to generate high-quality academic content. This study aims to evaluate the academic writing performance of both Qwen 2.5 Max and DeepSeek v3 by comparing these models with popular systems such as ChatGPT, Gemini, Llama, Mistral, and Gemma. In this research, 40 articles on the topics of Digital Twin and Healthcare were used. The method of this study involves using generative AI tools to generate texts based on posed questions and paraphrased abstracts of these 40 articles. Then, the generated texts were evaluated through the plagiarism tool, AI detection tools, word count comparisons, semantic similarity tools and readability assessments. It was observed that plagiarism test result rates were generally higher for the paraphrased abstract texts and lower for the answers generated to the questions, but both were above acceptable levels. In the evaluations made with the AI detection tool, it was determined with high accuracy that all the generated texts were detected as AI-generated. In terms of the generated word count comparison, it was evaluated that all chatbots generated satisfactory amount of content. Semantic similarity tests show that the generated texts have high semantic overlap with the original texts. The readability tests indicated that the generated texts were not sufficiently readable. © 2025, CC BY.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Leveraging Generalizability of Image-to-Image Translation for Enhanced Adversarial Defense

arXiv

引用

arXiv 2025年

作者： Zhang, Haibo Yao, Zhihua Sakurai, Kouichi Saitoh, Takeshi Department of Artificial Intelligence Faculty of Computer Science and Systems Engineering Kyushu Institute of Technology Japan Faculty of Economics and Business Administration The University of Kitakyushu Japan Department of Informatics Faculty of Information Science and Electrical Engineering Kyushu University Japan

In the rapidly evolving field of artificial intelligence, machine learning emerges as a key technology characterized by its vast potential and inherent risks. The stability and reliability of these models are important, as they are frequent targets of security threats. Adversarial attacks, first rigorously defined by Ian Goodfellow et al. in 2013, highlight a critical vulnerability: they can trick machine learning models into making incorrect predictions by applying nearly invisible perturbations to images. Although many studies have focused on constructing sophisticated defensive mechanisms to mitigate such attacks, they often overlook the substantial time and computational costs of training and maintaining these models. Ideally, a defense method should be able to generalize across various, even unseen, adversarial attacks with minimal overhead. Building on our previous work on image-to-image translation-based defenses, this study introduces an improved model that incorporates residual blocks to enhance generalizability. The proposed method requires training only a single model, effectively defends against diverse attack types, and is well-transferable between different target models. Experiments show that our model can restore the classification accuracy from near zero to an average of 72% while maintaining competitive performance compared to state-of-the-art methods. Significantly, our model operates more efficiently, reducing the time needed to process individual images and speeding up the training process to achieve faster convergence. Robustness tests further confirm stable performance under varying attack strengths, demonstrating the model’s practical value in real-world adversarial settings. Copyright © 2025, The Authors. All rights reserved.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

Knowledge Graph-based Retrieval-Augmented Generation for Schema Matching

arXiv

引用

arXiv 2025年

作者： Ma, Chuangtao Chakrabarti, Sriom Khan, Arijit Molnár, Bálint The Department of Computer Science Aalborg University Aalborg Denmark The Department of Information Systems Faculty of Informatics Eötvös Loránd University Budapest Hungary

Traditional similarity-based schema matching methods are incapable of resolving semantic ambiguities and conflicts in domain-specific complex mapping scenarios due to missing commonsense and domain-specific knowledge. The hallucination problem of large language models (LLMs) also makes it challenging for LLM-based schema matching to address the above issues. Therefore, we propose a Knowledge Graph-based Retrieval-Augmented Generation model for Schema Matching, referred to as the KG-RAG4SM. In particular, KG-RAG4SM introduces novel vector-based, graph traversal-based, and query-based graph retrievals, as well as a hybrid approach and ranking schemes that identify the most relevant subgraphs from external large knowledge graphs (KGs). We showcase that KG-based retrieval-augmented LLMs are capable of generating more accurate results for complex matching cases without any re-training. Our experimental results show that KG-RAG4SM outperforms the LLM-based state-of-the-art (SOTA) methods (e.g., Jellyfish-8B) by 35.89% and 30.50% in terms of precision and F1 score on the MIMIC dataset, respectively;KG-RAG4SM with GPT-4o-mini outperforms the pre-trained language model (PLM)-based SOTA methods (e.g., SMAT) by 69.20% and 21.97% in terms of precision and F1 score on the Synthea dataset, respectively. The results also demonstrate that our approach is more efficient in end-to-end schema matching, and scales to retrieve from large KGs. Our case studies on the dataset from the real-world schema matching scenario exhibit that the hallucination problem of LLMs for schema matching is well mitigated by our solution. Copyright © 2025, The Authors. All rights reserved.

关键词： Knowledge graph

来源：评论

学校读者我要写书评

暂无评论

Real-Time Manipulation Action Recognition with a Factorized Graph Sequence Encoder

arXiv

引用

arXiv 2025年

作者： Erdogan, Enes Aksoy, Eren Erdal Sariel, Sanem Artificial Intelligence and Robotics Lab Faculty of Computer and Informatics Engineering Istanbul Technical University Istanbul Turkey School of Information Technology Halmstad University Halmstad Sweden

Recognition of human manipulation actions in real-time is essential for safe and effective human-robot interaction and collaboration. The challenge lies in developing a model that is both lightweight enough for real-time execution and capable of generalization. While some existing methods in the literature can run in real-time, they struggle with temporal scalability, i.e., they fail to adapt to long-duration manipulations effectively. To address this, leveraging the generalizable scene graph representations, we propose a new Factorized Graph Sequence Encoder network that not only runs in real-time but also scales effectively in the temporal dimension, thanks to its factorized encoder architecture. Additionally, we introduce Hand Pooling operation, a simple pooling operation for more focused extraction of the graph-level embeddings. Our model outperforms the previous state-of-the-art real-time approach, achieving a 14.3% and 5.6% improvement in F1-macro score on the KIT Bimanual Action (Bimacs) Dataset and Collaborative Action (CoAx) Dataset, respectively. Moreover, we conduct an extensive ablation study to validate our network design choices. Finally, we compare our model with its architecturally similar RGB-based model on the Bimacs dataset and show the limitations of this model in contrast to ours on such an object-centric manipulation dataset. Copyright © 2025, The Authors. All rights reserved.

关键词： Human robot interaction

来源：评论

学校读者我要写书评

暂无评论

Open and Closed Source Models for LLM-Generated Metaheuristics Solving Engineering Optimization Problem 28th

Open and Closed Source Models for LLM-Generated Metaheuris...

引用

28th European Conference on Applications of Evolutionary Computation, EvoApplications 2025, held as part of EvoStar 2025

作者： Senkerik, Roman Viktorin, Adam Kadavy, Tomas Kovac, Jozef Janku, Peter Pekar, Libor Guzowski, Hubert Smolka, Maciej Byrski, Aleksander Pluhacek, Michal A.I.Lab Faculty of Applied Informatics Tomas Bata University Zlin Czech Republic Department of Automation and Control Engineering Faculty of Applied Informatics Tomas Bata University Zlin Czech Republic Faculty of Computer Science AGH University of Krakow Krakow Poland Center of Excellence in Artificial Intelligence AGH University of Krakow Krakow Poland

ISBN: (纸本)9783031900648

This paper explores the applicability of generative AI (genAI), specifically Large Language Models (LLMs), for the automatic generation and configuration of metaheuristic algorithms to address a real-world engineering problem: the optimal parameter estimation of time-delay systems in interconnected heating-cooling loops. The study introduces a pioneering workflow and iterative architecture with feedback within the emerging field of genAI-driven optimization for real optimization problems, eliminating the need for manually crafted or modified algorithms. This automated system empowers domain experts in engineering to solve complex optimization problems with minimal knowledge of optimization algorithms, lowering the barrier to entry for sophisticated algorithm use. We demonstrate how LLMs can generate effective optimizers under conditions like connstrained optimization problems where the solution lies near the boundaries of the search space. Four state-of-the-art LLMs (closed and open-sourced) have been selected for experiments. These are GPT-4o, GPT-4o mini, Claude Sonnet 3.5 and Llama 3.1. All studied LLMs generated metaheuristics that outperformed the initialization baseline optimization method (Random Search and CMA-ES). Notably, the Claude Sonnet 3.5 model generated a metaheuristic with the best mean results, almost matching the performance of the tuned state-of-the-art DISH algorithm, as an example of adaptive Differential Evolution. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Time of arrival

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：