检索结果-内蒙古大学图书馆

Align is not Enough: Multimodal Universal Jailbreak Attack against Multimodal Large Language Models

IEEE Transactions on Circuits and Systems for Video technology 2025年第6期35卷 5475-5488页

作者： Wang, Youze Hu, Wenbo Dong, Yinpeng Liu, Jing Zhang, Hanwang Hong, Richang Hefei University of Technology School of Computer Science and Information Engineering Hefei230009 China Tsinghua University Beijing100084 China Chinese academy of science Institute of automation Beijing100190 China Nanyang Technological University School of Computer Science and Engineering 639798 Singapore

Large Language Models (LLMs) have evolved into Multimodal Large Language Models (MLLMs), significantly enhancing their capabilities by integrating visual information and other types, thus aligning more closely with the nature of human intelligence, which processes a variety of data forms beyond just text. Despite advancements, the undesirable generation of these models remains a critical concern, particularly due to vulnerabilities exposed by text-based jailbreak attacks, which have represented a significant threat by challenging existing safety protocols. Motivated by the unique security risks posed by the integration of new and old modalities for MLLMs, we propose a unified multimodal universal jailbreak attack framework that leverages iterative image-text interactions and transfer-based strategy to generate a universal adversarial suffix and image. Our work not only highlights the interaction of image-text modalities can be used as a critical vulnerability but also validates that multimodal universal jailbreak attacks can bring higher-quality undesirable generations across different MLLMs. We evaluate the undesirable context generation of MLLMs like LLaVA, Yi-VL, MiniGPT4, MiniGPT-v2, and InstructBLIP, and reveal significant multimodal safety alignment issues, highlighting the inadequacy of current safety mechanisms against sophisticated multimodal attacks. This study underscores the urgent need for robust safety measures in MLLMs, advocating for a comprehensive review and enhancement of security protocols to mitigate potential risks associated with multimodal capabilities. © 1991-2012 IEEE.

关键词： Human form models

来源：评论

学校读者我要写书评

暂无评论

Cross domain recommendation using dual inductive transfer learning

引用

Multimedia Tools and Applications 2024年 1-25页

作者： Zaman, Nilufar Jana, Angshuman Computer Engineering Trinity Academy of Engineering Kondhwa-Saswad Road Maharashtra Pune411048 India Computer Science Indian Institute of Information Technology Guwahati Bongora Assam Guwahati781015 India

Cross Domain Recommendations have made a great impact on the field of online services. It helps the service provider of one domain to understand their users from the information of other domains to recommend the correct product to them. Unlike other traditional approaches, our approach finds both way latent relations between users and items considering the interactions between users and items. Traditional model normally avoids considering any implicit information about the user or item for calculating the latent relation between them. But our model Dual Inductive Transfer Learning(DITL) uses all possible user and item features and the ratings provided for recommendations. Our model learns the user’s preferences across the source and target domain and thus preserves the user’s taste across different but related domains. Our model auto-encodes the features in a feature vector to extract the user’s taste and thus transfers this across domains for efficient recommendation. The proposed model is tested on three huge benchmark datasets, the Douban dataset, the Amazon 5-crore dataset for the domain of movies, books and music whereas Tenrec dataset for the domain of articles and videos and it proves to outperform the classical cross-domain recommendation systems. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Recommender systems

来源：评论

学校读者我要写书评

暂无评论

Smartphone based human activity recognition irrespective of usage behavior using deep learning technique

引用

International Journal of Information technology (Singapore) 2025年第1期17卷 69-85页

作者： Kundu, Soumya Mallik, Manjarini Saha, Jayita Chowdhury, Chandreyee Department of Computer Science and Engineering Jadavpur University Kolkata India Department of Computer Science and Engineering Manipal Institute of Technology Bengaluru Manipal Academy of Higher Education Manipal India

Human activity recognition (HAR) from sensory data is a crucial task for a wide variety of applications. The in-built inertial sensor facilities of commercial smartphones have made the data collection process easier. However, different smartphone configurations exhibit variations in sensor readings for the same activities. Different smartphone holding positions, like in hand, shirt, or trouser pockets, also lead to variations in signal patterns for the same activity. Some recent works have shown that automated feature extraction using deep learning methods can significantly improve activity recognition, although there is a lack of experimentation considering device heterogeneity and different smartphone holding positions. The proposed work addresses this research gap with a two-fold contribution. First, a CNN-based HAR framework is proposed that forms 2-D frequency domain images to capture temporal patterns in the data along with inter-axis spatial features. Second, an ensemble of conditional classifiers has been designed based on CNN that exhibits generality in terms of device configurations and usage behavior. Real life data have been collected for different activities using different devices for experimentation. The proposed ensemble model is found to recognize activities with 94% accuracy even when the training and test devices are different for real datasets. © The Author(s) 2024.

关键词： Activity image CNN Deep learning Human activity recognition Smartphone

来源：评论

学校读者我要写书评

暂无评论

Skyway:Accelerate Graph Applications with a Dual-Path Architecture and Fine-Grained Data Management

引用

Journal of computer science & technology 2024年第4期39卷 871-894页

作者： Mo Zou Ming-Zhe Zhang Ru-Jia Wang Xian-He Sun Xiao-Chun Ye Dong-Rui Fan Zhi-Min Tang Institute of Computing Technology Chinese Academy of SciencesBeijing 100190China University of Chinese Academy of Sciences Beijing 100049China Institute of Information Engineering Chinese Academy of SciencesBeijing 100045China Department of Computer Science Illinois Institute of TechnologyChicagoIL 60616U.S.A.

Graph processing is a vital component of many AI and big data ***,due to its poor locality and complex data access patterns,graph processing is also a known performance killer of AI and big data *** this work,we propose to enhance graph processing applications by leveraging fine-grained memory access patterns with a dual-path architecture on top of existing software-based graph *** first identify that memory accesses to the offset,edge,and state array have distinct locality and impact on *** then introduce the Skyway architecture,which consists of two primary components:1)a dedicated direct data path between the core and memory to transfer state array elements efficiently,and 2)a data-type aware fine-grained memory-side row buffer hardware for both the newly designed direct data path and the regular memory hierarchy data *** proposed Skyway architecture is able to improve the overall performance by reducing the memory access interference and improving data access efficiency with a minimal *** evaluate Skyway on a set of diverse algorithms using large real-world *** a simulated fourcore system,Skyway improves the performance by 23%on average over the best-performing graph-specialized hardware optimizations.

关键词： graph application computer architecture memory hierarchy

来源：评论

学校读者我要写书评

暂无评论

"affectAlchemy": An Affective Dataset Based on Plutchik's Psychological Model for Text-Based Emotion Recognition and its Analysis Using ML Techniques 8

"affectAlchemy": An Affective Dataset Based on Plutchik's Ps...

引用

8th IEEE International Conference on Computing, Communication, Control and Automation, ICCUBEA 2024

作者： Kapase, Ajay Uke, Nilesh Savant, Jatin Desai, Mrunali Ghatage, Shravani Rahangdale, Atul Vishwakarma Institute of Information Technology Computer Science & Engineering Pune India Trinity Academy of Engineering Computer Science & Engineering Pune India KITs College of Engineering Computer Science & Engineering Kolhapur India

ISBN: (纸本)9798350391770

A technique to identify people's attitudes, and sentiments towards specified targets such as things, services, and subjects, is called sentiment analysis. As a dedicated subset of NLP, it deals with predicting specific reactions broadly characterizing sentiments as positive, negative, or neutral. However, the accuracy and depth vary since there are a variety of human emotions that exist rather than just positive, negative, and neutral. In the past few decades, the term affective computing has become well-liked which involves the use of AI for identifying, interpreting, and responding to human emotions. A part of it called text-based affective systems are artificial emotional intelligent systems that often categorize the text into core wide range of emotions like fear, sadness, anger, and happiness. These affective systems use various psychological models for emotion recognition. Our study incorporates the development of a dataset based on Robert Plutchik's psychological model of emotions, which sets forth eight primary emotions which are combined to form more complex emotions, resulting in a total of 24 secondary emotions resulting total of 32 variety of emotions. The work introduces the "AffectAlchemy"dataset which is validated using multiclass classification machine learning techniques. This dataset can contribute to the field of affective computing for the development of models that can do enhanced emotion recognition, and improved emotional context understanding to develop emotion-aware applications and facilitating research in affective systems. © 2024 IEEE.

关键词： Intelligent systems

来源：评论

学校读者我要写书评

暂无评论

Machine Learning for the Exposure of Malicious URLs 1

Machine Learning for the Exposure of Malicious URLs

引用

1st International Conference on Recent Advances in Information technology for Sustainable Development, ICRAIS 2023

作者： Hanji, Bhagyashri R. Kanagavalli, R. Dayananda Sagar Academy of Technology and Management Computer Science and Engineering Bangalore India Global Academy of Technology Computer Science and Engineering Bangalore India

ISBN: (纸本)9798350306637

In today's era, as per the statistics for world-wide internet users or digital population as on April 2023, there are 5.18 billion internet users worldwide, which goes to 64.6 percent of global population. Of this total, 4.8 billion, or 59.9 percent of the world's population are social media users. [***/statistics617136/digital-population-worldwide]. Enhancement in technology and internet usage increases the scope for cyber-criminal activities. As the connectivity and the number of users grow, there is a proportional increase in attackers. Vindictive sites or URLs become one of the significant dangers for network protection. Malevolent URLs are most risky danger prompting phishing assault, malware dissemination, Drive-by Downloads, Program Exploits, Data fraud in this computerized world. It's critical to remain cautious and safeguard us on the web. It is vital to be sufficiently mindful and have the option to identify and act against these dangers by not visiting such destinations or giving any private data. In this work we will focus on attacks through malicious URLS, working, preventions and machine learning algorithms used to detect malicious URLs. © 2023 IEEE.

关键词： Cybersecurity

来源：评论

学校读者我要写书评

暂无评论

3TFL-XLnet-CP: A Novel Transformer-Based Crop Yield Prediction Framework with Weighted Loss Based 3-Tier Feature Learning Model

引用

SN computer science 2025年第3期6卷 1-17页

作者： Anoop, G.L. Nandini, C. Naresh, E. Department of Computer Science and Engineering Christ University Karnataka Bengaluru India Department of Computer Science and Engineering Dayananda Sagar Academy of Technology and Management Karnataka Bengaluru India Department of Information Technology Manipal Institute of Technology Bengaluru Manipal Academy of Higher Education Manipal India

The advancement of crop yield prediction through artificial intelligence (AI) has gained significant attention. However, the existing AI-based approaches for maximizing agricultural productivity, specifically in crop yield prediction, have not consistently delivered satisfactory results. In response to this challenge, we propose a novel framework named as Three Tier Feature Learning with XLnet based Crop Prediction (3TFL-XLnet-CP) that enhances agricultural productivity by accurately predicting crop yield. The 3TFL-XLnet-CP framework employs a three-tier feature learning approach in combination with the powerful XLnet transformer-based crop prediction model. The three-tier feature learning involves the integration of Spiking Neural Network (SNN), Graphical Neural Network (GNN), and Convolutional Neural Network (CNN) to extract distinct feature vectors from the preprocessed data. These feature vectors are then concatenated using Jaccard Similarity to measure their similarity score. Additionally, a weighted Loss function is introduced to optimize feature learning, further enhanced by a novel self-adaptive Spider Monkey Optimization algorithm (SASMO). The concatenated features are subsequently fed into the classification layer for making precise crop yield predictions. The proposed model is implemented using the Python platform and evaluated against existing models such as ANN, RNN, DNN, and BiLSTM. The comparison demonstrates the superiority of our proposed 3TFL-XLnet-CP framework in accurately predicting crop yield. © The Author(s) 2025.

关键词： 3TFL-XLnet-CP Convolutional neural network Crop yield prediction Graphical neural network Spiking neural network

来源：评论

学校读者我要写书评

暂无评论

Machine Learning Models for Predicting Hypothyroidism: Utilizing Synthetic Data for Improved Accuracy

Machine Learning Models for Predicting Hypothyroidism: Utili...

引用

2024 Asia Pacific Conference on Innovation in technology, APCIT 2024

作者： Sajjan, Vinutha N Varsha, S. Sheela, S. Global Academy of Technology Department of Computer Science and Engineering Karnataka Bengaluru India

ISBN: (纸本)9798350361537

If left untreated, hypothyroidism, a common endocrine illness, can result in a number of health issues due to the underactive thyroid gland. Accurately predicting hypothyroidism is essential for prompt diagnosis and successful treatment of the illness. This work addresses class imbalance in hypothyroid datasets with and without SMOTE, examining the predictive power of the SVM, LR, RF, KNN and XgBoost algorithms. Models are assessed and trained using performance metrics that include F1-score, accuracy, precision, recall, and area under the ROC curve (AUC) using a dataset that includes clinical characteristics and thyroid function tests. According to the study's findings, all three models show excellent accuracy. Random Forest performs better than SVM and LR, especially when SMOTE is included, and provides better predictive capacities for the diagnosis of *** work emphasizes the need of addressing class imbalance in medical predictive modeling tasks and offers useful information for the diagnosis and treatment of thyroid diseases. © 2024 IEEE.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Optimized Convolutional Neural Network at the IoT edge for image detection using pruning and quantization

引用

Multimedia Tools and Applications 2025年第9期84卷 5435-5455页

作者： Naveen, Soumyalatha Kounte, Manjunath R Department of Computer Science and Engineering Manipal Institute of Technology Bengaluru Manipal Academy of Higher Education Manipal India School of Electronics and Communication Engineering REVA University Bangalore India

Most real-time computer vision applications heavily rely on Convolutional Neural Network (CNN) based models, for image classification and recognition. Due to the computationally and memory-intensive nature of the CNN model, it’s challenging to deploy on resource-constrained Internet of Things (IoT) devices to enable Edge intelligence for real-time decision-making. Edge intelligence requires minimum inference latency, memory footprint, and energy-efficient model. This work aims to develop an energy-efficient deep learning accelerator using a 3-stage pipeline: Training, Weight-pruning, and Quantization to reduce the model size and optimize the resources. First, we employ YOLOv3, a CNN architecture to detect objects in an image on the trained data. In addition, a sparse network of YOLO has been created by using pruning, which helps to improve the network’s performance and efficiency by reducing the computational requirements. Finally, we utilize 8-bit quantization to reduce the precision of the weights and activations, in a neural network. The evaluation of our proposed model shows that combining pruning and 8-bit quantization improves the efficiency and performance of the model. While pruning shows a decline of 80.39% in model parameters. The combination of 8-bit quantization results in an improvement in inference latency by 22.72% compared to existing SQuantization approach and a reduction of energy consumption by 29.41%. © The Author(s) 2024.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

The Evolution of Large Language Model: Models, Applications and Challenges 12

The Evolution of Large Language Model: Models, Applications ...

引用

12th International Conference on Current Trends in Advanced Computing, ICCTAC 2024

作者： Sindhu, B. Prathamesh, R.P. Sameera, M.B. KumaraSwamy, S. Global Academy of Technology Department of Computer Science and Engineering Karnataka Bangalore India

ISBN: (数字)9798350395808

ISBN: (纸本)9798350395808

Large Language Models (LLMs) have attracted a lot of attention due to their success in natural language processing tasks. This paper provides a thorough overview by examining the architecture, applications, problems, assessment techniques, and future directions of LLM. With the constantly growing body of literature, a succinct yet comprehensive overview of recent developments is essential. Following the development of NLP, it highlights the move from rule-based systems to sophisticated transformer structures like as BERT and GPT. Important LLMs for text creation, translation, and summarization are mentioned, including T5, BART, and BioGPT. LLM performance is evaluated using metrics including as accuracy, perplexity, BLEU score, and ROUGE score. Research is still being done because of issues with bias, overfitting, and real-Time processing. Future directions include managing longer contexts, lowering bias, and increasing efficiency through methods like federated learning. Continuous learning and multimodal LLMs are promising fields, as well as interpretive AI. In conclusion, LLMs have transformed natural language processing (NLP) and brought up both technical and ethical issues about the future of AI. © 2024 IEEE.

关键词： Natural language processing systems

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：