检索结果-内蒙古大学图书馆

The superalignment of superhuman intelligence with large language models

science China(Information sciences) 2025年第6期68卷 101-111页

作者： Minlie HUANG Yingkang WANG Shiyao CUI Pei KE Jie TANG The CoAI Group Department of Computer Science and Technology Tsinghua University Laboratory of Intelligent Collaborative Computing University of Electronic Science and Technology of China Knowledge Engineering Group Department of Computer Science and Technology Tsinghua University

We have witnessed the emergence of superhuman intelligence thanks to the fast development of large language models(LLMs) and multimodal language models. As the application of such superhuman models becomes increasingly popular, a critical question arises: how can we ensure they still remain safe, reliable, and aligned well with human values encompassing moral values, Schwartz's Values, ethics, and many more? In this position paper, we discuss the concept of superalignment from a learning perspective to answer this question by outlining the learning paradigm shift from large-scale pretraining and supervised fine-tuning, to alignment training. We define superalignment as designing effective and efficient alignment algorithms to learn from noisy-labeled data(point-wise samples or pair-wise preference data) in a scalable way when the task is very complex for human experts to annotate and when the model is stronger than human experts. We highlight some key research problems in superalignment, namely, weak-to-strong generalization, scalable oversight, and evaluation. We then present a conceptual framework for superalignment, which comprises three modules: an attacker which generates the adversary queries trying to expose the weaknesses of a learner model, a learner which refines itself by learning from scalable feedbacks generated by a critic model with minimal human experts, and a critic which generates critics or explanations for a given query-response pair, with a target of improving the learner by criticizing. We discuss some important research problems in each component of this framework and highlight some interesting research ideas that are closely related to our proposed framework, for instance, self-alignment, self-play, self-refinement, and more. Last, we highlight some future research directions for superalignment, including the identification of new emergent risks and multi-dimensional alignment.

关键词： superalignment superhuman intelligence large language models scalable feedback weak-to-strong generalization

来源：评论

学校读者我要写书评

暂无评论

A survey on cross-user federated recommendation

引用

science China(Information sciences) 2025年第4期68卷 7-32页

作者： Enyue YANG Yudi XIONG Wei YUAN Weike PAN Qiang YANG Zhong MING College of Computer Science and Software Engineering Shenzhen University School of Electrical Engineering and Computer Science The University of Queensland WeBank AI Lab WeBank Department of Computer Science and Engineering Hong Kong University of Science and Technology College of Big Data and Internet Shenzhen Technology University Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ)

Recommender systems are effective in mitigating information overload, yet the centralized storage of user data raises significant privacy concerns. Cross-user federated recommendation(CUFR) provides a promising distributed paradigm to address these concerns by enabling privacy-preserving recommendations directly on user devices. In this survey, we review and categorize current progress in CUFR, focusing on four key aspects: privacy, security, accuracy, and efficiency. Firstly,we conduct an in-depth privacy analysis, discuss various cases of privacy leakage, and then review recent methods for privacy protection. Secondly, we analyze security concerns and review recent methods for untargeted and targeted *** untargeted attack methods, we categorize them into data poisoning attack methods and parameter poisoning attack methods. For targeted attack methods, we categorize them into user-based methods and item-based methods. Thirdly,we provide an overview of the federated variants of some representative methods, and then review the recent methods for improving accuracy from two categories: data heterogeneity and high-order information. Fourthly, we review recent methods for improving training efficiency from two categories: client sampling and model compression. Finally, we conclude this survey and explore some potential future research topics in CUFR.

关键词： cross-user federated recommendation federated recommendation federated learning recommender systems user privacy

来源：评论

学校读者我要写书评

暂无评论

WiFo: wireless foundation model for channel prediction

引用

science China(Information sciences) 2025年第6期68卷 372-384页

作者： Boxun LIU Shijian GAO Xuanyu LIU Xiang CHENG Liuqing YANG State Key Laboratory of Photonics and Communications School of Electronics Peking University Internet of Things Thrust The Hong Kong University of Science and Technology (Guangzhou) Department of Electronic and Computer Engineering and Department of Civil and Environmental Engineering The Hong Kong University of Science and Technology

Channel prediction permits to acquire channel state information(CSI) without signaling overhead. However,almost all existing channel prediction methods necessitate the deployment of a dedicated model to accommodate a specific configuration. Leveraging the powerful modeling and multi-task learning capabilities of foundation models, we propose the first space-time-frequency(STF) wireless foundation model(WiFo) to address time-frequency channel prediction tasks in a unified manner. Specifically, WiFo is initially pre-trained over massive and extensive diverse CSI datasets. Then, the model will be instantly used for channel prediction under various CSI configurations without any fine-tuning. We propose a masked autoencoder(MAE)-based network structure for WiFo to handle heterogeneous STF CSI data, and design several mask reconstruction tasks for self-supervised pre-training to capture the inherent 3D variations of CSI. To fully unleash its predictive power, we build a large-scale heterogeneous simulated CSI dataset consisting of 160k CSI samples for *** validate its superior unified learning performance across multiple datasets and demonstrate its state-of-the-art(SOTA) zero-shot generalization performance via comparisons with other full-shot baselines.

关键词： channel prediction channel state information foundation model self-supervised pre-training zero-shot learning

来源：评论

学校读者我要写书评

暂无评论

Committed-programming reductions: formalizations,implications and relations

引用

science China(Information sciences) 2024年第10期67卷 151-171页

作者： Jiang ZHANG Yu YU Dengguo FENG Shuqin FAN Zhenfeng ZHANG State Key Laboratory of Cryptology Department of Computer Science and Engineering Shanghai Jiao Tong University Trusted Computing and Information Assurance Laboratory Institute of SoftwareChinese Academy of Sciences

In this work, we introduce a class of black-box(BB) reductions called committed-programming reduction(CPRed) in the random oracle model(ROM) and obtain the following interesting results:(1) we demonstrate that some well-known schemes, including the full-domain hash(FDH) signature(Eurocrypt1996) and the Boneh-Franklin identity-based encryption(IBE) scheme(Crypto 2001), are provably secure under CPReds;(2) we prove that a CPRed associated with an instance-extraction algorithm implies a reduction in the quantum ROM(QROM). This unifies several recent results, including the security of the Gentry-Peikert-Vaikuntanathan IBE scheme by Zhandry(Crypto 2012) and the key encapsulation mechanism(KEM) variants using the Fujisaki-Okamoto transform by Jiang et al.(Crypto 2018) in the ***, we show that CPReds are incomparable to non-programming reductions(NPReds) and randomly-programming reductions(RPReds) formalized by Fischlin et al.(Asiacrypt 2010).

关键词： provable security random oracle model quantum random oracle model black-box reduction/separation programmability

来源：评论

学校读者我要写书评

暂无评论

P3DC:Reducing DRAM Cache Hit Latency by Hybrid Mappings

引用

Journal of computer science & Technology 2024年第6期39卷 1341-1360页

作者： Ye Chi Ren-Tong Guo Xiao-Fei Liao Hai-Kun Liu Jianhui Yue National Engineering Research Center for Big Data Technology and System Wuhan 430074China Services Computing Technology and System Laboratory Wuhan 430074China Cluster and Grid Computing Laboratory Wuhan 430074China School of Computer Science and Technology Huazhong University of Science and TechnologyWuhan 430074China School of Big Data and Internet Shenzhen Technology UniversityShenzhen 518118China Department of Computer Science Michigan Technological UniversityHoughton 49931-1295U.S.A.

Die-stacked dynamic random access memory(DRAM)caches are increasingly advocated to bridge the performance gap between the on-chip cache and the main *** fully realize their potential,it is essential to improve DRAM cache hit rate and lower its cache hit *** order to take advantage of the high hit-rate of set-association and the low hit latency of direct-mapping at the same time,we propose a partial direct-mapped die-stacked DRAM cache called *** design is motivated by a key observation,i.e.,applying a unified mapping policy to different types of blocks cannot achieve a high cache hit rate and low hit latency *** address this problem,P3DC classifies data blocks into leading blocks and following blocks,and places them at static positions and dynamic positions,respectively,in a unified set-associative *** also propose a replacement policy to balance the miss penalty and the temporal locality of different *** addition,P3DC provides a policy to mitigate cache thrashing due to block type *** results demonstrate that P3DC can reduce the cache hit latency by 20.5%while achieving a similar cache hit rate compared with typical set-associative caches.P3DC improves the instructions per cycle(IPC)by up to 66%(12%on average)compared with the state-of-the-art direct-mapped cache—BEAR,and by up to 19%(6%on average)compared with the tag-data decoupled set-associative cache—DEC-A8.

关键词： die-stacked dynamic random access memory(DRAM) cache set-associative direct-mapped hit latency

来源：评论

学校读者我要写书评

暂无评论

Nonlinear Filtering With Sample-Based Approximation Under Constrained Communication:Progress, Insights and Trends

引用

IEEE/CAA Journal of Automatica Sinica 2024年第7期11卷 1539-1556页

作者： Weihao Song Zidong Wang Zhongkui Li Jianan Wang Qing-Long Han IEEE the State Key Laboratory for Turbulence and Complex Systems Department of Mechanics and Engineering ScienceCollege of Engineering Peking University the Department of Computer Science Brunel University London the School of Aerospace Engineering Beijing Institute of Technology the School of Science Computing and Engineering Technologies Swinburne University of Technology

The nonlinear filtering problem has enduringly been an active research topic in both academia and industry due to its ever-growing theoretical importance and practical *** main objective of nonlinear filtering is to infer the states of a nonlinear dynamical system of interest based on the available noisy measurements. In recent years, the advance of network communication technology has not only popularized the networked systems with apparent advantages in terms of installation,cost and maintenance, but also brought about a series of challenges to the design of nonlinear filtering algorithms, among which the communication constraint has been recognized as a dominating concern. In this context, a great number of investigations have been launched towards the networked nonlinear filtering problem with communication constraints, and many samplebased nonlinear filters have been developed to deal with the highly nonlinear and/or non-Gaussian scenarios. The aim of this paper is to provide a timely survey about the recent advances on the sample-based networked nonlinear filtering problem from the perspective of communication constraints. More specifically, we first review three important families of sample-based filtering methods known as the unscented Kalman filter, particle filter,and maximum correntropy filter. Then, the latest developments are surveyed with stress on the topics regarding incomplete/imperfect information, limited resources and cyber ***, several challenges and open problems are highlighted to shed some lights on the possible trends of future research in this realm.

关键词： Communication constraints maximum correntropy filter networked nonlinear filtering particle filter sample-based approximation unscented Kalman filter

来源：评论

学校读者我要写书评

暂无评论

Digital Twins and Cyber-Physical Systems:A New Frontier in computer Modeling

引用

computer Modeling in engineering & sciences 2025年第4期143卷 51-113页

作者： Vidyalakshmi G S Gopikrishnan Wadii Boulila Anis Koubaa Gautam Srivastava Department of Computer Science and Engineering SRM University APAmaravati522240India School of Computer Science and Engineering VIT-AP UniversityAmaravathi522240India Robotics and Internet-of-Things Laboratory Prince Sultan UniversityRiyadh12435Saudi Arabia Department of Math and Computer Science Brandon UniversityBrandonMB R7A6A9Canada Research Centre for Interneural Computing China Medical UniversityTaichung40402Taiwan

Cyber-Physical Systems(CPS)represent an integration of computational and physical elements,revolutionizing industries by enabling real-time monitoring,control,and optimization.A complementary technology,Digital Twin(DT),acts as a virtual replica of physical assets or processes,facilitating better decision making through simulations and predictive *** and DT underpin the evolution of Industry 4.0 by bridging the physical and digital *** survey explores their synergy,highlighting how DT enriches CPS with dynamic modeling,realtime data integration,and advanced simulation *** layered architecture of DTs within CPS is examined,showcasing the enabling technologies and tools vital for seamless *** study addresses key challenges in CPS modeling,such as concurrency and communication,and underscores the importance of DT in overcoming these *** in various sectors are analyzed,including smart manufacturing,healthcare,and urban planning,emphasizing the transformative potential of CPS-DT *** addition,the review identifies gaps in existing methodologies and proposes future research directions to develop comprehensive,scalable,and secure CPSDT *** synthesizing insights fromthe current literature and presenting a taxonomy of CPS and DT,this survey serves as a foundational reference for academics and *** findings stress the need for unified frameworks that align CPS and DT with emerging technologies,fostering innovation and efficiency in the digital transformation era.

关键词： Cyber physical systems digital twin efficiency Industry 4.0 robustness and intelligence

来源：评论

学校读者我要写书评

暂无评论

Covid-19 Classification using Fine-tuned EfficientNet Architecture 9

Covid-19 Classification using Fine-tuned EfficientNet Archit...

引用

9th IEEE International Conference for Convergence in Technology, I2CT 2024

作者： Sruthi, Srigiri Emadaboina, Siddharth MacHavarapu, Pradyumna Singh, Rimjhim Padam Kanchan, Sneha Amrita Vishwa Vidyapeetham Amrita School of Computing Department of Computer Science And Engineering Bengaluru India Universiti Tunku Abdul Rahman Department of Internet Engineering And Computer Science Malaysia

ISBN: (纸本)9798350394474

This work utilizes an Efficient Net-based deep learning architecture, leveraging its efficiency and scalability for image classification tasks. We employ a Dataset of chest X-ray images, including COVID-19 cases as well as other relevant Pathology, to train and evaluate the model. The training process involves data augmentation, model optimization, and early stopping to ensure robust and generalizable performance. It specifies the image size, channels, and class count, and then proceeds to create a pre-trained model using the EfficientNetB0 architecture. The model is built upon the pre-trained EfficientNetB0 model, with additional layers including Layer Normalization, Dense, and Dropout layers for fine-tuning on the Covid-19 data. Subsequently, the model is rigorously evaluated using standard performance metrics, including precision, recall, F1-score, and accuracy, and has proven to outperform other state-of-art models by a 3% increase in F1-score value. © 2024 IEEE.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

A Machine Learning Approach for Credit Card Fraud Detection in Massive Datasets Using SMOTE and Random Sampling 6

A Machine Learning Approach for Credit Card Fraud Detection ...

引用

6th IEEE International Conference on Recent Advances in Intelligent Computational Systems, RAICS 2024

作者： Sreenivas Prasad, B. Akash Babu, N. Reddy, Harthikeswar Singh, Rimjhim Padam Kanchan, Sneha Amrita School of Computing Department of Computer Science & Engineering Amrita Vishwa Vidyapeetham Bengaluru India Universiti Tunku Abdul Rahman Department of Internet Engineering & Computer science Malaysia

ISBN: (纸本)9798350381689

The surge in digital transactions has paved the way for an alarming rise in credit card fraud, compelling the need for robust detection systems. The swift progress of technology has transformed customer payment habits, driving them toward a cashless society, with the rise of digital payments making it easier for fraudsters to commit fraud. The impact of credit card fraud on revenue can be significant. No matter the scale of the businesses, credit card fraud will have a significant effect on the businesses. This paper addresses the escalating threat by proposing a fraud detection model utilizing the 'IEEE CIS Credit Fraud Detection' dataset from Kaggle, constructing a machine learning model aimed at enhancing the accuracy of fraudulent transaction identification, presenting an integrated approach to credit card fraud detection that combines user separation and innovative techniques for improved accuracy and real-world impact, ultimately strengthening the global financial ecosystem. © 2024 IEEE.

关键词： Electronic money

来源：评论

学校读者我要写书评

暂无评论

Automation of Text Summarization using Hugging Face NLP 5

Automation of Text Summarization using Hugging Face NLP

引用

5th IEEE International Conference for Emerging Technology, INCET 2024

作者： Asmitha, M. Danda, Aashritha Bysani, Hemanth Singh, Rimjhim Padam Kanchan, Sneha Department of Computer Science and Engineering Amrita School of Computing Amrita Vishwa Vidyapeetham Bengaluru India Department of Internet Engineering and Computer science Universiti Tunku Abdul Rahman Malaysia

ISBN: (纸本)9798350361155

Within the expansive domain of"Natural Language Processing" (NLP), the task of"text summarization" emerges as a foundational element, playing a pivotal role in distilling relevant information from extensive textual corpora. In the digital age, the importance of efficient summarization becomes increasingly critical, given the overwhelming volume of textual information. This comprehensive study delves into the intricacies of both extractive and abstractive summarization techniques, placing a specific focus on transformer-based models like BERT and GPT. These models, celebrated for their remarkable capabilities in context comprehension and coherent summarization, are rigorously evaluated alongside established methods like TF-IDF, TextRank, Sumy, Fine Tuning Transformers, Model-T5, LSTM, greedy, and beam search. The practical implications of text summarization extend across diverse fields, encompassing news stories, academic papers, and social media content, underscoring its broad utility in various domains. This study not only incorporates cutting-edge models but also explores a gamut of evaluation methods to discern the quality of summarization. By intertwining theory and application, this research positions itself at the forefront of evolving summarization approaches, shedding light on the transformative impact on information consumption patterns. The dynamic landscape of summarization methods underscores the need for continuous research and innovation, as technological advancements continue to reshape how individuals access and comprehend information. © 2024 IEEE.

关键词： Quality control

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

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

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：