检索结果-南通市图书馆

维普期刊数据库

Skill enhancement learning with knowledge distillation

在线全文

学校读者我要写书评

暂无评论

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

SDCC: software-defined collective communication for distributed training

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

science China(Information sciences) 2024年第9期67卷 105-125页

作者： Xin JIN Zhen ZHANG Yunshan JIA Yun MA Xuanzhe LIU School of computer science Peking University department.of computer science and technology Johns Hopkins University

Communication is crucial to the performance of distributed training. Today's solutions tightly couple the control and data planes and lack flexibility, generality, and performance. In this study, we present SDCC, a software-defined collective communication framework for distributed training. SDCC is based on the principle of modern systems design to effectively decouple the control plane from the data *** abstracts the operations for collective communication in distributed training with dataflow operations and unifies computing and communication with a single dataflow graph. The abstraction, together with the unification, is powerful: it enables users to easily express new and existing collective communication algorithms and optimizations, simplifies the integration with different computing engines(e.g., Py Torch and Tensor Flow) and network transports(e.g., Linux TCP and kernel bypass), and allows the system to improve performance by exploiting parallelism exposed by the dataflow graph. We further demonstrate the benefits of SDCC in four use cases.

关键词： machine learning systems distributed training deep learning collective communication software-defined networking

Memory-enhanced text style transfer with dynamic style learning and calibration

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

science China(Information sciences) 2024年第4期67卷 181-196页

作者： Fuqiang LIN Yiping SONG Zhiliang TIAN Wangqun CHEN Diwen DONG Bo LIU College of computer science and technology National University of Defense Technology department.of computer science and Engineering The Hong Kong University of Science and Technology Strategic Assessments and Consultation Institute Academy of Military Sciences

Text style transfer aims to rephrase a sentence to match the desired style while retaining the original content. As a controllable text generation task, mainstream approaches use content-independent style embedding as control variables to guide stylistic generation. Nonetheless, stylistic properties are contextsensitive even under the same style. For example, “delicious” and “helpful” convey positive sentiments,although they are more likely to describe food and people, respectively. Therefore, desired style signals must vary with the content. To this end, we propose a memory-enhanced transfer method, which learns fine-grained style representation concerning content to assist transfer. Rather than employing static style embedding or latent variables, our method abstracts linguistic characteristics from training corpora and memorizes subdivided content with the corresponding style representations. The style signal is dynamically retrieved from memory using the content as a query, providing a more expressive and flexible latent style space. To address the imbalance between quantity and quality in different content, we further introduce a calibration method to augment memory construction by modeling the relationship between candidate *** results obtained using three benchmark datasets confirm the superior performance of our model compared to competitive approaches. The evaluation metrics and case study also indicate that our model can generate diverse stylistic phrases matching context.

关键词： style transfer memory-enhanced method text generation deep learning text representation

Federated Learning Security and Privacy-Preserving Algorithm and Experiments Research Under Internet of Things Critical Infrastructure

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Tsinghua science and technology 2024年第2期29卷 400-414页

作者： Nasir Ahmad Jalali Hongsong Chen department.of computer science University of Science and Technology Beijing(USTB)Beijing 100083China

The widespread use of the Internet of Things(IoTs)and the rapid development of artificial intelligence technologies have enabled applications to cross commercial and industrial band *** such systems,all participants related to commercial and industrial systems must communicate and generate ***,due to the small storage capacities of IoT devices,they are required to store and transfer the generated data to third-party entity called“cloud”,which creates one single point to store their ***,as the number of participants increases,the size of generated data also ***,such a centralized mechanism for data collection and exchange between participants is likely to face numerous challenges in terms of security,privacy,and *** address these challenges,Federated Learning(FL)has been proposed as a reasonable decentralizing approach,in which clients no longer need to transfer and store real data in the central ***,they only share updated training models that are trained over their private *** the same time,FL enables clients in distributed systems to share their machine learning models collaboratively without their training data,thus reducing data privacy and security ***,slow model training and the execution of additional unnecessary communication rounds may hinder FL applications from operating properly in a distributed ***,these unnecessary communication rounds make the system vulnerable to security and privacy issues,because irrelevant model updates are sent between clients and ***,in this work,we propose an algorithm for fully homomorphic encryption called Cheon-Kim-Kim-Song(CKKS)to encrypt model parameters for their local information privacy-preserving *** proposed solution uses the impetus term to speed up model convergence during the model training ***,it establishes a secure communication channel between IoT devices and the *** a

关键词： Federated Learning(FL) Internet of Things(IoTs) lightweight Transport Layer Security(iTLS) Cheon-Kim-Kim-Song(CKKS)

Unbalanced private set intersection with linear communication complexity

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

science China(Information sciences) 2024年第3期67卷 79-93页

作者： Quanyu ZHAO Bingbing JIANG Yuan ZHANG Heng WANG Yunlong MAO Sheng ZHONG department.of computer science and technology Nanjing University Purple Mountain Lab

The private set intersection(PSI) protocol allows two parties holding a set of integers to compute the intersection of their sets without revealing any additional information to each other. The unbalanced PSI schemes consider a specific setting where a client holds a small set of the size n and a server holds a much larger set of the size m(n ■ m). The communication overhead of state-of-the-art balanced PSI schemes is O(m + n) and the unbalanced PSI schemes are O(nlogm). In this paper, we propose a novel secure unbalanced PSI protocol based on a hash proof system. The communication complexity of our protocol grows only linearly with the size of the small set. In other words, our protocol achieves communication overhead of O(n). We test the performance on a personal computer(PC) machine with a local area network(LAN)setting for the network. The experimental results demonstrate that the client only takes 2.01 s of online computation, 4.27 MB of round trip communication to intersect 1600 pieces of 32-bit integers with 220pieces of 32-bit integers with the security parameter λ = 512. Our protocol is efficient and can be applied to resource-constrained devices, such as cell phones.

关键词： unbalanced private set intersection Hash proof system linear communication complexity small set resource-constrained devices

Self-Aligning Multi-Modal Transformer for Oropharyngeal Swab Point Localization

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Tsinghua science and technology 2024年第4期29卷 1082-1091页

作者： Tianyu Liu Fuchun Sun department.of computer science and technology Tsinghua UniversityBeijing 100083China

The oropharyngeal swabbing is a pre-diagnostic procedure used to test various respiratory diseases, including COVID and Influenza A (H1N1). To improve the testing efficiency of testing, a real-time, accurate, and robust sampling point localization algorithm is needed for robots. However, current solutions rely heavily on visual input, which is not reliable enough for large-scale deployment. The transformer has significantly improved the performance of image-related tasks and challenged the dominance of traditional convolutional neural networks (CNNs) in the image field. Inspired by its success, we propose a novel self-aligning multi-modal transformer (SAMMT) to dynamically attend to different parts of unaligned feature maps, preventing information loss caused by perspective disparity and simplifying overall implementation. Unlike preexisting multi-modal transformers, our attention mechanism works in image space instead of embedding space, rendering the need for the sensor registration process obsolete. To facilitate the multi-modal task, we collected and annotate an oropharynx localization/segmentation dataset by trained medical personnel. This dataset is open-sourced and can be used for future multi-modal research. Our experiments show that our model improves the performance of the localization task by 4.2% compared to the pure visual model, and reduces the pixel-wise error rate of the segmentation task by 16.7% compared to the CNN baseline.

关键词： segmentation localization transformer multi-modal perception robotic perception

CToMP: a cycle-task-oriented memory protection scheme for unmanned systems

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

science China(Information sciences) 2024年第6期67卷 399-417页

作者： Chengyan MA Ning XI Di LU Yebo FENG Jianfeng MA School of Cyber Engineering Xidian University School of computer science and technology Xidian University department.of computer and Information science University of Oregon

Memory corruption attacks(MCAs) refer to malicious behaviors of system intruders that modify the contents of a memory location to disrupt the normal operation of computing systems, causing leakage of sensitive data or perturbations to ongoing processes. Unlike general-purpose systems, unmanned systems cannot deploy complete security protection schemes, due to their limitations in size, cost and *** in unmanned systems are particularly difficult to defend against. Furthermore, MCAs have diverse and unpredictable attack interfaces in unmanned systems, severely impacting digital and physical sectors. In this paper, we first generalize, model and taxonomize MCAs found in unmanned systems currently, laying the foundation for designing a portable and general defense approach. According to different attack mechanisms,we found that MCAs are mainly categorized into two types — return2libc and return2shellcode. To tackle return2libc attacks, we model the erratic operation of unmanned systems with cycles and then propose a cycle-task-oriented memory protection(CToMP) approach to protect control flows from tampering. To defend against return2shellcode attacks, we introduce a secure process stack with a randomized memory address by leveraging the memory pool to prevent Shellcode from being executed. Moreover, we discuss the mechanism by which CTo MP resists the return-oriented programming(ROP) attack, a novel variant of return2libc attacks. Finally, we implement CTo MP on CUAV V5+ with Ardupilot and Crazyflie. The evaluation and security analysis results demonstrate that the proposed approach CTo MP is resilient to various MCAs in unmanned systems with low footprints and system overhead.

关键词： unmanned system memory corruption attack memory protection system security randomized memory address

Fast Cross-Platform Binary Code Similarity Detection Framework Based on CFGs Taking Advantage of NLP and Inductive GNN

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Chinese Journal of Electronics 2024年第1期33卷 128-138页

作者： Jinxue PENG Yong WANG Jingfeng XUE Zhenyan LIU School of computer science and technology Beijing Institute of Technology

Cross-platform binary code similarity detection aims at detecting whether two or more pieces of binary code are similar or not. Existing approaches that combine control flow graphs(CFGs)-based function representation and graph convolutional network(GCN)-based similarity analysis are the best-performing ones. Due to a large amount of convolutional computation and the loss of structural information, the use of convolution networks will inevitably bring problems such as high overhead and sometimes inaccuracy. To address these issues, we propose a fast cross-platform binary code similarity detection framework that takes advantage of natural language processing(NLP)and inductive graph neural network(GNN) for basic blocks embedding and function representation respectively by simulating extracting structural features and temporal features. GNN’s node-centric and small batch is a suitable training way for large CFGs, it can greatly reduce computational overhead. Various NLP basic block embedding models and GNNs are evaluated. Experimental results show that the scheme with long short term memory(LSTM)for basic blocks embedding and inductive learning-based Graph SAGE(GAE) for function representation outperforms the state-of-the-art works. In our framework, we can take only 45% overhead. Improve efficiency significantly with a small performance trade-off.

关键词： Training Semantics Process control Binary codes Feature extraction Natural language processing Graph neural networks

同方期刊数据库

DIR:Dynamic Request Interleaving for Improving the Read Performance of Aged Solid-State Drives

在线全文

学校读者我要写书评

暂无评论

Journal of computer science & technology 2024年第1期39卷 82-98页

作者：聂世强张驰伍卫国 department.of computer science and technology Xi'an Jiaotong UniversityShaanci 710049China

Triple-level cell(TLC)NAND flash is increasingly adopted to build solid-state drives(SSDs)for modern computer *** TLC NAND flash effectively improves storage density,it faces severe reliability issues;in partic-ular,the pages exhibit different raw bit error rates(RBERs).Integrating strong low-density parity-check(LDPC)code helps to improve reliability but suffers from prolonged and proportional read latency due to multiple read retries for worse *** straightforward idea is that dispersing page-size data across several pages in different types can achieve a low-er average RBER and reduce the read ***,directly implementing this simple idea into flash translation lay-er(FTL)induces the read amplification issue as one logic page residing in more than one physical page brings several read *** this paper,we propose the Dynamic Request Interleaving(DIR)technology for improving the performance of TLC NAND flash-based SSDs,in particular,the aged ones with large *** exploits the observation that the la-tency of an I/O request is determined,without considering the queuing time,by the access of the slowest device page,i.e.,the page that has the highest *** grouping consecutive logical pages that have high locality and interleaving their encoded data in different types of device pages that have different RBERs,DIR effectively reduces the number of read re-tries for LDPC with limited read *** meet the requirement of allocating hybrid page types for interleaved data,we also design a page-interleaving friendly page allocation scheme,which splits all the planes into multi-plane re-gions for storing the interleaved data and single-plane regions for storing the normal *** pages in the multi-plane re-gion can be read/written in parallel by the proposed multi-plane command and avoid the read amplification *** on the DIR scheme and the proposed page allocation scheme,we build DIR-enable FTL,which integrates the pro

关键词： triple-layer cell solid-state drive(TLC SSD) performance interleaving data unbalanced bit error rate