检索结果-内蒙古大学图书馆

Video2Reward: Generating Reward Function from Videos for Legged Robot Behavior Learning

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zeng, Runhao Zhou, Dingjie Liang, Qiwei Liu, Junlin Li, Hui Huang, Changxin Li, Jianqiang Hu, Xiping Sun, Fuchun Artificial Intelligence Research Institute Shenzhen MSU-BIT University China College of Mechatronics and Control Engineering Shenzhen University China College of Computer Science and Software Engineering Shenzhen University China National Engineering Laboratory for Big Data System Computing Technology Shenzhen University China Department of Computer Science and Technology Tsinghua University China

Learning behavior in legged robots presents a significant challenge due to its inherent instability and complex constraints. Recent research has proposed the use of a large language model (LLM) to generate reward functions in reinforcement learning, thereby replacing the need for manually designed rewards by experts. However, this approach, which relies on textual descriptions to define learning objectives, fails to achieve controllable and precise behavior learning with clear directionality. In this paper, we introduce a new video2reward method, which directly generates reward functions from videos depicting the behaviors to be mimicked and learned. Specifically, we first process videos containing the target behaviors, converting the motion information of individuals in the videos into keypoint trajectories represented as coordinates through a video2text transforming module. These trajectories are then fed into an LLM to generate the reward function, which in turn is used to train the policy. To enhance the quality of the reward function, we develop a video-assisted iterative reward refinement scheme that visually assesses the learned behaviors and provides textual feedback to the LLM. This feedback guides the LLM to continually refine the reward function, ultimately facilitating more efficient behavior learning. Experimental results on tasks involving bipedal and quadrupedal robot motion control demonstrate that our method surpasses the performance of state-of-the-art LLM-based reward generation methods by over 37.6% in terms of human normalized score. More importantly, by switching video inputs, we find our method can rapidly learn diverse motion behaviors such as walking and running. Copyright © 2024, The Authors. All rights reserved.

关键词： Robots

Energy Efficient Design of Active STAR-RIS-Aided SWIPT Systems

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Faramarzi, Sajad Zarini, Hosein Javadi, Sepideh Mili, Mohammad Robat Zhang, Rui Karagiannidis, George K. Al-Dhahir, Naofal The School of Electrical Engineering Iran University of Science & Technology Tehran Iran The Dept. of Computer Engineering Sharif University of Technology Tehran Iran Tehran Iran School of Science and Engineering Shenzhen Research Institute of Big Data The Chinese University of Hong Kong Guangdong Shenzhen518172 China The Department of Electrical and Computer Engineering National University of Singapore Singapore117583 Singapore Department of Electrical and Computer Engineering Aristotle University of Thessaloniki Greece Lebanon The Department of Electrical and Computer Engineering The University of Texas at Dallas RichardsonTX75080 United States

In this paper, we consider the downlink transmission of a multi-antenna base station (BS) supported by an active simultaneously transmitting and reconfigurable intelligent surface (STAR-RIS) to serve single-antenna users via simultaneous wireless information and power transfer (SWIPT). In this context, we formulate an energy efficiency maximisation problem that jointly optimises the gain, element selection and phase shift matrices of the active STAR-RIS, the transmit beamforming of the BS and the power splitting ratio of the users. With respect to the highly coupled and non-convex form of this problem, an alternating optimisation solution approach is proposed, using tools from convex optimisation and reinforcement learning. Specifically, semi-definite relaxation (SDR), difference of concave functions (DC), and fractional programming techniques are employed to transform the non-convex optimisation problem into a convex form for optimising the BS beamforming vector and the power splitting ratio of the SWIPT. Then, by integrating meta-learning with the modified deep deterministic policy gradient (DDPG) and soft actor-critical (SAC) methods, a combinatorial reinforcement learning network is developed to optimise the element selection, gain and phase shift matrices of the active STAR-RIS. Our simulations show the effectiveness of the proposed resource allocation scheme. Furthermore, our proposed active STAR-RIS-based SWIPT system outperforms its passive counterpart by 57% on average. © 2024, CC BY.

关键词： Convex optimization

Multi-objective Aerial Collaborative Secure Communication Optimization via Generative Diffusion Model-enabled Deep Reinforcement Learning

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zhang, Chuang Sun, Geng Li, Jiahui Wu, Qingqing Wang, Jiacheng Niyato, Dusit Liu, Yuanwei The College of Computer Science and Technology Jilin University Changchun130012 China The Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University Changchun130012 China The College of Computer Science and Technology Jilin University Changchun130012 China The College of Computing and Data Science Nanyang Technological University 639798 Singapore The Department of Electronic Engineering Shanghai Jiao Tong University Shanghai China The College of Computing and Data Science Nanyang Technological University 639798 Singapore The School of Electronic Engineering and Computer Science Queen Mary University of London LondonE1 4NS United Kingdom

Due to flexibility and low-cost, unmanned aerial vehicles (UAVs) are increasingly crucial for enhancing coverage and functionality of wireless networks. However, incorporating UAVs into next-generation wireless communication systems poses significant challenges, particularly in sustaining high-rate and long-range secure communications against eavesdropping attacks. In this work, we consider a UAV swarm-enabled secure surveillance network system, where a UAV swarm forms a virtual antenna array to transmit sensitive surveillance data to a remote base station (RBS) via collaborative beamforming (CB) so as to resist mobile eavesdroppers. Specifically, we formulate an aerial secure communication and energy efficiency multi-objective optimization problem (ASCEE-MOP) to maximize the secrecy rate of the system and to minimize the flight energy consumption of the UAV swarm. To address the non-convex, NP-hard and dynamic ASCEE-MOP, we propose a generative diffusion model-enabled twin delayed deep deterministic policy gradient (GDMTD3) method. Specifically, GDMTD3 leverages an innovative application of diffusion models to determine optimal excitation current weights and position decisions of UAVs. The diffusion models can better capture the complex dynamics and the trade-off of the ASCEE-MOP, thereby yielding promising solutions. Simulation results highlight the superior performance of the proposed approach compared with traditional deployment strategies and some other deep reinforcement learning (DRL) benchmarks. Moreover, performance analysis under various parameter settings of GDMTD3 and different numbers of UAVs verifies the robustness of the proposed approach. Copyright © 2024, The Authors. All rights reserved.

关键词： Energy efficiency

Efficient Video Summarization with Hydra Attentive Vision Transformer

学校读者我要写书评

暂无评论

Efficient Video Summarization with Hydra Attentive Vision Tr...

Frontiers of Information Technology (FIT)

作者： Muhammad Shafqat Ali Muhammad Azhar Saba Masood Bumshik Lee Tanzeela Iqbal Adeen Amjad Department of Computer Science Comsats University Islamabad Sahiwal Pakistan Department of Applied Data Science Hong Kong Shue Yan University Hong Kong Department of Information and Communication Engineering Chosun University Gwangju Republic of Korea

A video summary’s objective is to produce succinct and condensed synopses that accurately depict the content of the original video while sacrificing none of its vital features. Effective deep summarization models have arisen in the field of video summarizing, made feasible by the advancement of gated recursive unit (GRU) and long and short-term memory (LSTM) technologies. However, if the video is very long, GRU and LSTM models are unlikely to capture long-term dependencies as well as they could otherwise. In recent years, significant progress in the field of supervised video summarization has been accomplished through the use of techniques involving sequence-to-sequence learning. However, it is important to remember that traditional recurrent neural networks (RNNs) have some limitations when it comes to modeling long sequences and that using transformers to represent sequences necessitates a large number of input parameters, resulting in increased computational complexity. We present a new video summarizing methodology that addresses the aforementioned issues by utilizing a Hydra Attention-based Vision Transformer framework. The suggested method captures long-term dependencies and extracts significant characteristics from video sequences well. The proposed architecture improves the accuracy and reduces the processing time of video summaries by using the capabilities of Hydra Attention and transformer-based sequence modeling. Our solution outperforms state-of-the-art approaches in terms of performance and computing economy in empirical evaluations of the SumMe and TVSum datasets.

关键词：

Transition between distribution patterns in human dynamics with high activity

学校读者我要写书评

暂无评论

Physical Review Research 2023年第3期5卷 033179-033179页

作者： Ming Zhao Jiani Chen Jiasheng Lao Yanqing Hu Jiarong Xie College of Physical Science and Technology Guangxi Normal University Guilin 541004 People's Republic of China Department of Statistics and Data Science College of Science Southern University of Science and Technology Shenzhen 518055 People's Republic of China School of Computer Science and Engineering Sun Yat-sen University Guangzhou 510006 People's Republic of China

Complex interactions among a large number of individuals lead to multiple patterns of collective human behavior. However, the theoretical prediction of pattern transitions has not been empirically confirmed. This is because in previous empirical studies, different patterns were observed in different systems, and the coexistence of multiple patterns in the same system was rarely found. By investigating nearly 10 million messages in 252 QQ groups, we find rich distribution patterns of interevent time for human collective behavior, including the transitions between bimodal distribution, double-power-law distribution, and single-power-law distribution. The model developed in this paper suggests that human physiological rhythms and the high collective activity play key roles in the presentation of the single-power-law distribution. These results enhance the empirical research in the field of human dynamics, and are helpful for understanding many complex socioeconomic phenomena.

关键词： Collective behavior in networks Stochastic processes Social dynamics Time series analysis

Image-Based Soil Health Analysis Using Deep Learning for Precision Agriculture

学校读者我要写书评

暂无评论

Image-Based Soil Health Analysis Using Deep Learning for Pre...

International Conference on Communication and Electronics Systems (ICCES)

作者： V. S. Saranya Saikiran Mangali K Srinija Galeiah Medabalimi Meena Devi R Venkata Ramana N Ajanthaa Lakkshmanan Department of Computing Technologies School of Computing SRM Institute of Science and Technology Kattankulathur Department of CSBS B V RAJU INSTITUTE OF TECHNOLOGY Medak Telangana Department: Data science MLRIT DUNDIGAL Sagi Ramakrishnam Raju Engineering College Sagi Ramakrishnam Raju Marg Bhimavaram Andhra Pradesh India Department of electrical and electronics engineering KCG college of technology Chennai Department of Computer Science and Engineering Koneru Lakshmaiah Education Foundation

ISBN: (数字)9798350377972

ISBN: (纸本)9798350377989

Soil health is a critical component of agricultural productivity and sustainability. Traditional methods of assessing soil health, including physical sampling and laboratory testing, are labor-intensive, time-consuming, and costly. This paper proposes a novel image-based approach for soil health analysis using deep learning techniques. By utilizing aerial images captured by drones or satellites equipped with multispectral and thermal sensors, the system non-invasively assesses key soil health indicators such as moisture content, nutrient levels, organic matter, and erosion patterns. The proposed method integrates image processing techniques with machine learning models, particularly Convolutional Neural Networks (CNNs), to extract relevant soil features from the images. These CNNs are trained to predict soil health metrics, and the system's performance is evaluated through accuracy, mean squared error (MSE), and F1-score, showing significant improvements over traditional methods. This scalable solution offers a real-time tool for precision agriculture, aiding farmers and agronomists in making informed, data-driven decisions regarding soil health management.

关键词： Precision agriculture Deep learning Moisture Soil Thermal sensors Nutrients Feature extraction Real-time systems Convolutional neural networks Monitoring

RedAgent: Red Teaming Large Language Models with Context-aware Autonomous Language Agent

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Xu, Huiyu Zhang, Wenhui Wang, Zhibo Xiao, Feng Zheng, Rui Feng, Yunhe Ba, Zhongjie Ren, Kui The State Key Laboratory of Blockchain and Data Security Zhejiang University China School of Cyber Science and Technology Zhejiang University China Palo Alto Networks United States Department of Computer Science and Engineering University of North Texas United States

Recently, advanced Large Language Models (LLMs) such as GPT-4 have been integrated into many real-world applications like Code Copilot. These applications have significantly expanded the attack surface of LLMs, exposing them to a variety of threats. Among them, jailbreak attacks that induce toxic responses through carefully constructed jailbreak prompts have raised critical safety concerns. To effectively identify these threats, a growing number of red team approaches simulate potential adversarial scenarios by crafting jailbreak prompts to test the target LLM. However, existing red teaming methods do not consider the unique vulnerabilities of LLM in different scenarios (e.g., code-related tasks), making it difficult to adjust the jailbreak prompts to find context-specific vulnerabilities, thereby lacking efficiency. Meanwhile, these methods are limited to refining handcrafted jailbreak templates using a few mutation operations (such as synonym replacement), lacking the automation and scalability to continuously adapt to different scenarios. To enable context-aware and efficient red teaming, we abstract and model existing attacks into a coherent concept called "jailbreak strategy" and propose a multi-agent LLM system named RedAgent that leverages these strategies to generate context-aware jailbreak prompts. By self-reflecting on contextual feedback and red teaming trials in an additional memory buffer, RedAgent continuously learns how to leverage these strategies to achieve more effective jailbreaks in specific contexts. Extensive experiments demonstrate that our system can jailbreak most black-box LLMs within just five queries, improving the efficiency of existing red teaming methods by two times. Additionally, RedAgent can jailbreak customized LLM applications more efficiently. By generating context-aware jailbreak prompts towards trending applications on GPTs, we discover 60 severe vulnerabilities of these real-world applications with only two queries per vulnera

关键词： Autonomous agents

Cybersecurity Threat Detection using Machine Learning on Network Traffic

学校读者我要写书评

暂无评论

Cybersecurity Threat Detection using Machine Learning on Net...

International Conference on ent Computing and Control Systems (ICICCS)

作者： R. Rithi N. Nithya A. Akila R. Saravanakumar M. Vargheese S. Jeya Lakshmi Master of Computer Applications Karpagam Academy of Higher Education Coimbatore Tamil Nadu India Department of Computer Applications Karpagam Academy of Higher Education Coimbatore Tamil Nadu India Department of Artificial Intelligence and Data Science R.M.K College of Engineering and Technology Tamil Nadu India Iconix Software Solution Tirunelveli Tamil Nadu India Department of Computer Science and Engineering PSN College of Engineering and Technology Tirunelveli Tamil Nadu India

ISBN: (数字)9798331512088

ISBN: (纸本)9798331512095

Web phishing is a prevalent online attack in which victims are tricked into revealing their personal information by visiting fraudulent websites. Phishers often use URL shortening tools to disguise malicious links, making it easier to steal sensitive data such as usernames, passwords, and banking details. These phishing websites are designed to mimic legitimate ones, making detection more challenging. As phishing techniques continue to evolve rapidly alongside technological advancements, the need for effective anti-phishing systems has become increasingly crucial. Machine learning has emerged as a powerful tool for recognizing and mitigating phishing scams. This article provides a comprehensive overview of machine learning techniques for phishing detection and their effectiveness in combating online threats.

关键词： Uniform resource locators Accuracy Reviews Phishing Scalability Machine learning Telecommunication traffic Passwords Threat assessment Real-time systems

SA-WavLM: Speaker-Aware Self-Supervised Pre-training for Mixture Speech

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Lin, Jingru Ge, Meng Ao, Junyi Deng, Liqun Li, Haizhou Department of Electrical and Computer Engineering National University of Singapore Singapore Huawei Noah's Ark Lab Hong Kong School of Data Science Shenzhen Research Institute of Big Data The Chinese University of Hong Kong Shenzhen China

It was shown that pre-trained models with self-supervised learning (SSL) techniques are effective in various downstream speech tasks. However, most such models are trained on single-speaker speech data, limiting their effectiveness in mixture speech. This motivates us to explore pre-training on mixture speech. This work presents SA-WavLM, a novel pre-trained model for mixture speech. Specifically, SA-WavLM follows an "extract-merge-predict" pipeline in which the representations of each speaker in the input mixture are first extracted individually and then merged before the final prediction. In this pipeline, SA-WavLM performs speaker-informed extractions with the consideration of the interactions between different speakers. Furthermore, a speaker shuffling strategy is proposed to enhance the robustness towards the speaker absence. Experiments show that SA-WavLM either matches or improves upon the state-of-the-art pre-trained models. Copyright © 2024, The Authors. All rights reserved.

关键词： Speech recognition