检索结果-南通市图书馆

Research on General-Purpose Brain-Inspired Computing Systems

Journal of computer science & technology 2024年第1期39卷 4-21页

作者：渠鹏纪兴龙陈嘉杰庞猛李宇晨刘晓义张悠慧 department of computer science and technology Tsinghua UniversityBeijing 100084China department of Precision Instrument Tsinghua UniversityBeijing 100084China

Brain-inspired computing is a new technology that draws on the principles of brain science and is oriented to the efficient development of artificial general intelligence(AGI),and a brain-inspired computing system is a hierarchical system composed of neuromorphic chips,basic software and hardware,and algorithms/applications that embody this *** the system is developing rapidly,it faces various challenges and opportunities brought by interdisciplinary research,including the issue of software and hardware *** paper analyzes the status quo of brain-inspired computing *** by some design principle and methodology of general-purpose computers,it is proposed to construct"general-purpose"brain-inspired computing systems.A general-purpose brain-inspired computing system refers to a brain-inspired computing hierarchy constructed based on the design philosophy of decoupling software and hardware,which can flexibly support various brain-inspired computing applications and neuromorphic chips with different ***,this paper introduces our recent work in these aspects,including the ANN(artificial neural network)/SNN(spiking neural network)development tools,the hardware agnostic compilation infrastructure,and the chip micro-archi-tecture with high flexibility of programming and high performance;these studies show that the"general-purpose"system can remarkably improve the efficiency of application development and enhance the productivity of basic software,thereby being conductive to accelerating the advancement of various brain-inspired algorithms and *** believe that this is the key to the collaborative research and development,and the evolution of applications,basic software and chips in this field,and conducive to building a favorable software/hardware ecosystem of brain-inspired computing.

关键词： brain-inspired computing neuromorphic chip compiler spiking neural network

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Graph Accelerators—A Case for Sparse Data Processing

引用

Journal of computer science & technology 2024年第2期39卷 243-244页

作者：陈文光 department of computer science and technology Tsinghua UniversityBeijing 100084China

Graph is a powerful sparse data structure that intuitively represents entities and their *** graph traversal algorithms such as Breadth-First Search(BFS),Single-Source Shortest Path(SSSP),PageRank,and Weakly Connected Components(WCC)have extensive applications in social network analysis,risk management for finance,and recommendation ***,graph processing in CPUs and GPUs is not very efficient due to its irregular memory *** people have proposed software approaches to speed up graph processing,such as PowerGraph,PowerLyra,and Shentu,which address load imbalance issues by replicating high-degree *** and GridGraph attempt to improve memory access locality by scanning the edge list of graphs while localizing the range of vertices accessed in a *** and Gemini provide adaptive dual compute modes(bottom-up and topdown),which are particularly effective for BFS-like algorithms such as BFS and ***,pure software approaches have their limitations,and it is desired to see how hardware could be employed to accelerate graph processing.

关键词： intuitive desired limitations

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

SWG: an architecture for sparse weight gradient computation

引用

science China(Information sciences) 2024年第2期67卷 302-321页

作者： Weiwei WU Fengbin TU Xiangyu LI Shaojun WEI Shouyi YIN School of Integrated Circuits Tsinghua University department of Electronic and computer Engineering The Hong Kong University of Science and Technology

On-device training for deep neural networks(DNN) has become a trend due to various user preferences and scenarios. The DNN training process consists of three phases, feedforward(FF), backpropagation(BP), and weight gradient(WG) update. WG takes about one-third of the computation in the whole training process. Current training accelerators usually ignore the special computation property of WG and process it in a way similar to FF/BP. Besides, the extensive data sparsity existing in WG, which brings opportunities to save computation, is not well explored. Nevertheless, exploiting the optimization opportunities would meet three underutilization problems, which are caused by(1) the mismatch between WG data dimensions and hardware parallelism,(2) the full sparsity, i.e., the sparsity of feature map(Fmap),error map(Emap), and gradient, and(3) the workload imbalance resulting from irregular sparsity. In this paper, we propose a specific architecture for sparse weight gradient(SWG) computation. The architecture is designed based on hierarchical unrolling and sparsity-aware(HUSA) dataflow to exploit the optimization opportunities of the special computation property and full data sparsity. In HUSA dataflow, the data dimensions are unrolled hierarchically on the hardware architecture. A valid-data trace(VDT) mechanism is embedded in the dataflow to avoid the underutilization caused by the two-sided input sparsity. The gradient is unrolled in PE to alleviate the underutilization induced by output sparsity while maintaining the data reuse opportunities. Besides, we design an intra-and inter-column balancer(IIBLC) to dynamically tackle the workload imbalance problem resulting from the irregular sparsity. Experimental results show that with HUSA dataflow exploiting the full sparsity, SWG achieves a speedup of 12.23× over state-of-the-art gradient computation architecture, Train Ware. SWG helps to improve the energy efficiency of the state-of-the-art training accelerator LNPU from

关键词： CNN training gradient computation sparsity architecture

来源：

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Understanding adversarial attacks on observations in deep reinforcement learning

引用

science China(Information sciences) 2024年第5期67卷 69-83页

作者： You QIAOBEN Chengyang YING Xinning ZHOU Hang SU Jun ZHU Bo ZHANG department of computer science and technology Beijing National Research Center for Information Science and Technology Tsinghua-Bosch Joint Center for Machine Learning Institute for Artificial Intelligence Tsinghua University Peng Cheng Laboratory

Deep reinforcement learning models are vulnerable to adversarial attacks that can decrease the cumulative expected reward of a victim by manipulating its observations. Despite the efficiency of previous optimization-based methods for generating adversarial noise in supervised learning, such methods might not achieve the lowest cumulative reward since they do not generally explore the environmental ***, a framework is provided to better understand the existing methods by reformulating the problem of adversarial attacks on reinforcement learning in the function space. The reformulation approach adopted herein generates an optimal adversary in the function space of targeted attacks, repelling them via a generic two-stage framework. In the first stage, a deceptive policy is trained by hacking the environment and discovering a set of trajectories routing to the lowest reward or the worst-case performance. Next, the adversary misleads the victim to imitate the deceptive policy by perturbing the observations. Compared to existing approaches, it is theoretically shown that our adversary is strong under an appropriate noise level. Extensive experiments demonstrate the superiority of the proposed method in terms of efficiency and effectiveness, achieving state-of-the-art performance in both Atari and MuJoCo environments.

关键词： deep learning reinforcement learning adversarial robustness adversarial attack

来源：

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

A smart hybrid memory scheduling approach using neural models

引用

science China(Information sciences) 2024年第4期67卷 333-334页

作者： Yanjie ZHEN Huijun ZHANG Yongheng DENG Weining CHEN Wei GAO Ju REN Yu CHEN department of computer science and technology Tsinghua University China Huaneng Clean Energy Research Institute

Traditional dynamic random access memory is limited by its low capacity and high cost because of the increasing demand for the memory capacity of applications. Hybrid memory architectures, including new memory hardwar... 详细信息

关键词： Cost reduction

来源：

同方期刊数据库评论

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Building a High-Performance Graph Storage on Top of Tree-Structured Key-Value Stores

引用

Big Data Mining and Analytics 2024年第1期7卷 156-170页

作者： Heng Lin Zhiyong Wang Shipeng Qi Xiaowei Zhu Chuntao Hong Wenguang Chen Yingwei Luo School of computer science Peking UniversityBeijing 100871China Ant Group Beijing 100020China department of computer science and technology Tsinghua UniversityBeijing 100084China

Graph databases have gained widespread adoption in various industries and have been utilized in a range of applications,including financial risk assessment,commodity recommendation,and data lineage *** the principles and design of these databases have been the subject of some investigation,there remains a lack of comprehensive examination of aspects such as storage layout,query language,and *** present study focuses on the design and implementation of graph storage layout,with a particular emphasis on tree-structured key-value *** also examine different design choices in the graph storage layer and present our findings through the development of TuGraph,a highly efficient single-machine graph database that significantly outperforms well-known Graph DataBase Management System(GDBMS).Additionally,TuGraph demonstrates superior performance in the Linked Data Benchmark Council(LDBC)Social Network Benchmark(SNB)interactive benchmark.

关键词： graph database high-performance graph storage

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Skill enhancement learning with knowledge distillation

引用

science China(Information sciences) 2024年第8期67卷 206-220页

作者： Naijun LIU Fuchun SUN Bin FANG Huaping LIU department of computer science and technology Tsinghua University

Skill learning through reinforcement learning has significantly progressed in recent years. However, it often struggles to efficiently find optimal or near-optimal policies due to the inherent trial-and-error exploration in reinforcement learning. Although algorithms have been proposed to enhance skill learning efficacy, there is still much room for improvement in terms of skill learning performance and training stability. In this paper, we propose an algorithm called skill enhancement learning with knowledge distillation(SELKD), which integrates multiple actors and multiple critics for skill learning. SELKD employs knowledge distillation to establish a mutual learning mechanism among actors. To mitigate critic overestimation bias,we introduce a novel target value calculation method. We also perform theoretical analysis to ensure the convergence of SELKD. Finally, experiments are conducted on several continuous control tasks, illustrating the effectiveness of the proposed algorithm.

关键词： skill learning enhancement learning reinforcement learning knowledge distillation

来源：

同方期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

The Application of Artificial Intelligence in Alzheimer’s Research

引用

tsinghua science and technology 2024年第1期29卷 13-33页

作者： Qing Zhao Hanrui Xu Jianqiang Li Faheem Akhtar Rajput Liyan Qiao Faculty of Information technology Beijing University of TechnologyBeijing 100124China School of Medicine Tsinghua UniversityBeijing 100084China department of computer science Sukkur IBA UniversitySukkur 65200Pakistan department of Neurology Tsinghua University Yuquan HospitalBeijing 100040China

Alzheimer's disease(AD)is an irreversible and neurodegenerative disease that slowly impairs memory and neurocognitive function,but the etiology of AD is still *** the explosive growth of electronic health data,the application of artificial inteligence(Al)in the healthcare setting provides excellent potential for exploring etiology and personalized treatment approaches,and improving the disease's diagnostic and prognostic *** paper first briefly introduces Al technologies and applications in medicine,and then presents a comprehensive review of Al in *** simple,it includes etiology discovery based on genetic data,computer-aided diagnosis(CAD),computer-aided prognosis(CAP)of AD using multi-modality data(genetic,neuroimaging and linguistic data),and pharmacological or non-pharmacological approaches for treating ***,some popular publicly available AD datasets are introduced,which are important for advancing Al technologies in AD ***,core research challenges and future research directions are discussed.

关键词： Alzheimer's disease artificial intelligence etiology discovery computer-aided diagnosis computer-aided prognosis treatment

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

A globally shared resource paradigm for encoded storage systems in the public cloud

引用

Fundamental Research 2024年第3期4卷 642-650页

作者： Zhiyue Li Guangyan Zhang department of computer science and technology Tsinghua UniversityBeijing 100084China Beijing National Research Center for Information science and technology Tsinghua UniversityBeijing 100084China

Public clouds favor sharing of storage resources,in which many tenants acquire bandwidth and storage capacity from a shared storage *** provide high availability,data are often encoded to provide fault tolerance with low storage *** this,efficiently organizing an encoded storage system for shared I/Os is critical for application *** is usually hard to achieve as different applications have different stripe configurations and fault tolerance *** this paper,we first study the block trace from the Alibaba cloud,and find that I/O patterns of modern applications prefer the resource sharing *** on this,we propose a globally shared resource paradigm for encoded storage system in the public *** globally shared resource paradigm can provide balanced load and fault tolerance for numerous disk pool sizes and arbitrary application stripe ***,we demonstrate with two case studies that our theory can help address the device-specific problems of HDD and SSD RAID arrays with slight modifications:comparing the existing resource partition and resource sharing methods,our theory can promote the rebuild speed of the HDD RAID arrays by 2.5,and reduce the P99 tail latency of the SSD arrays by up to two orders of magnitude.

关键词： Public cloud Encoded storage Load balancing RAID reconstruction Tail latency

来源：

维普期刊数据库评论

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

ChatGPT: potential, prospects, and limitations

引用

Frontiers of Information technology & Electronic Engineering 2024年第1期25卷 6-11页

作者： Jie ZHOU Pei KE Xipeng QIU Minlie HUANG Junping ZHANG School of computer science Fudan UniversityShanghai 200433China department of computer science and technology Tsinghua UniversityBeijing 100084China Shanghai Key Laboratory of Intelligent Information Processing Shanghai 200433China

Recently,OpenAI released Chat Generative Pre-trained Transformer(ChatGPT)(Schulman et al.,2022)(https://***),which has attracted considerable attention from the industry and academia because of its impressive *** is the first time that such a variety of open tasks can be well solved within one large language *** better understand ChatGPT,we briefly introduce its history,discuss its advantages and disadvantages,and point out several potential ***,we analyze its impact on the development of trustworthy artificial intelligence,conversational search engine,and artificial general intelligence.

关键词： artificial limitations intelligence

来源：

维普期刊数据库

同方期刊数据库评论

在线全文

学校读者我要写书评

暂无评论

欢迎您,

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

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

请选择保存的检索档案：

请选择收藏分类：

通借通还

欢迎您,

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

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

在线全文

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：