检索结果-内蒙古大学图书馆

Reinforcement learning-based traffic signal control

学校读者我要写书评

暂无评论

Reinforcement learning-based traffic signal control

International Conference on Emerging Trends in Networks and computer Communications (ETNCC)

作者： Liu Tian Sheng Afizan Bin Azman Navid Ali Khan Siva Raja Sindiramutty Suresh Singh Sahil Verma School of Computer Science Taylor's University Subang Jaya Selengor Malaysia Division of Research and Development Lovely Professional University Phagwara India KKR and KSR Institute of Technology and Sciences Guntur Andhra Pradesh India Department of Computer Science and Engineering Chandigarh Group of Colleges Jhanjeri Mohali Punjab India

ISBN: (数字)9798350353266

ISBN: (纸本)9798350353273

Traffic signal control (TSC) is an effective method for easing traffic congestion and enhancing traffic efficiency in urban areas, particularly with the growing urbanization. Conventional traffic signal control techniques are unable to adapt swiftly to the intricate and dynamic road environment, necessitating a more intelligent approach to signal control. In recent years, there has been an increasing interest in employing reinforcement learning (RL) for TSC, which has displayed considerable potential in optimizing control strategies for complex traffic conditions. This article focuses on the application of RL in TSC for research purposes. To recreate real urban traffic conditions, a traffic simulation system was used in this study. Extensive experiments and comparative studies conducted in a simulated environment have illustrated the effectiveness and superiority of reinforcement learning methods in traffic signal control. The article primarily employs Deep Q-Network (DQN) in reinforcement learning to build the model and examines the variations in traffic signal timing under different reward mechanisms. In conclusion, this study delves into traffic signal adaptation strategies, conducts experiments that integrate the virtual and real world, and effectively addresses urban traffic congestion issues. The research findings demonstrate that reinforcement learning can significantly impact traffic signal control, effectively resolving current traffic signal control challenges and advancing the development of urban road network traffic efficiency.

关键词： Roads Urban areas Transportation Reinforcement learning Telecommunication traffic Traffic control Timing Safety Traffic congestion Signal resolution

STEER: Unified Style Transfer with Expert Reinforcement

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Hallinan, Skyler Brahman, Faeze Lu, Ximing Jung, Jaehun Welleck, Sean Choi, Yejin Paul G. Allen School of Computer Science & Engineering University of Washington United States Allen Institute for AI Language Technologies Institute Carnegie Mellon University United States

While text style transfer has many applications across natural language processing, the core premise of transferring from a single source style is unrealistic in a real-world setting. In this work, we focus on arbitrary style transfer: rewriting a text from an arbitrary, unknown style to a target style. We propose STEER: Unified Style Transfer with Expert Reinforcement, a unified framework developed to overcome the challenge of limited parallel data for style transfer. STEER involves automatically generating a corpus of style-transfer pairs using a product of experts during decoding. The generated offline data is then used to pre-train an initial policy before switching to online, off-policy reinforcement learning for further improvements via fine-grained reward signals. STEER is unified and can transfer to multiple target styles from an arbitrary, unknown source style, making it particularly flexible and efficient. Experimental results on a challenging dataset with text from a diverse set of styles demonstrate state-of-the-art results compared to competitive baselines. Remarkably, STEER outperforms the 175B parameter instruction-tuned GPT-3 on overall style transfer quality, despite being 226 times smaller in size. We also show STEER is robust, maintaining its style transfer capabilities on out-of-domain data, and surpassing nearly all baselines across various styles. The success of our method highlights the potential of RL algorithms when augmented with controllable decoding to overcome the challenge of limited data supervision. Copyright © 2023, The Authors. All rights reserved.

关键词： Decoding

Fine-grained Retrieval Prompt Tuning

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Wang, Shijie Chang, Jianlong Wang, Zhihui Li, Haojie Ouyang, Wanli Tian, Qi International School of Information Science & Engineering Dalian University of Technology China Huawei Cloud & AI China College of Computer and Engineering Shandong University of Science and Technology China SenseTime Computer Vision Research Group The University of Sydney Australia

Fine-grained object retrieval aims to learn discriminative representation to retrieve visually similar objects. However, existing top-performing works usually impose pairwise similarities on the semantic embedding spaces or design a localization sub-network to continually fine-tune the entire model in limited data scenarios, thus resulting in convergence to suboptimal solutions. In this paper, we develop Fine-grained Retrieval Prompt Tuning (FRPT), which steers a frozen pre-trained model to perform the fine-grained retrieval task from the perspectives of sample prompting and feature adaptation. Specifically, FRPT only needs to learn fewer parameters in the prompt and adaptation instead of fine-tuning the entire model, thus solving the issue of convergence to suboptimal solutions caused by fine-tuning the entire model. Technically, a discriminative perturbation prompt (DPP) is introduced and deemed as a sample prompting process, which amplifies and even exaggerates some discriminative elements contributing to category prediction via a content-aware inhomogeneous sampling operation. In this way, DPP can make the fine-grained retrieval task aided by the perturbation prompts close to the solved task during the original pre-training. Thereby, it preserves the generalization and discrimination of representation extracted from input samples. Besides, a category-specific awareness head is proposed and regarded as feature adaptation, which removes the species discrepancies in features extracted by the pre-trained model using category-guided instance normalization. And thus, it makes the optimized features only include the discrepancies among subcategories. Extensive experiments demonstrate that our FRPT with fewer learnable parameters achieves the state-of-the-art performance on three widely-used fine-grained datasets. © 2022, CC0.

关键词： Semantics

Online Implementation of the Local Eigenvalue Modification Procedure for High-Rate Model Assimilation 41st

学校读者我要写书评

暂无评论

Online Implementation of the Local Eigenvalue Modification P...

41st IMAC, A Conference and Exposition on Structural Dynamics, 2023

作者： Vereen, Alexander B. Ogunniyi, Emmanuel A. Downey, Austin R. J. Dodson, Jacob Moura, Adriane G. Bakos, Jason D. Department of Mechanical Engineering University of South Carolina ColumbiaSC United States Department of Civil and Environmental Engineering University of South Carolina ColumbiaSC United States Air Force Research Laboratory Munitions Directorate Eglin Air Force Base EglinFL United States Applied Research Associates Emerald Coast Division NicevilleFL United States Department of Computer Science and Engineering University of South Carolina ColumbiaSC United States

ISBN: (纸本)9783031349416

High-rate structural health monitoring of active structures operating in high-rate dynamic environments empowers the execution of preventative behavior in response to structural degradation or external stimuli. Examples of structures operating in high-rate dynamic environments include hypersonic vehicles, space crafts, and ballistic packages. The effective selection of reactive actions to be taken in real time requires an up-to-date model of the structure’s state. Importantly, the short timescale of relevance to these structures means that the model must be continuously updated with a time step of 1 millisecond or less. However, traditional frequency-based methods for updating the finite element model online require solving the generalized eigenvalue problem, which becomes more complex as the number of nodes or FEA model increases, thereby increasing computational time. In this work, the local eigenvalue modification procedure is put forward to accelerate the extraction of natural frequencies from finite element models updated online. The local eigenvalue modification procedure works by precomputing the eigenvalue solution to a reference state of the system and then computes the single (i.e., local) change in the modal domain from the reference state to the current state online. The modal domain update in the local eigenvalue modification procedure bypasses the general eigenvalue problem, which is the most expensive computational step. For the online implementation of the state estimation, the Dynamic Reproduction of Projectiles in Ballistic Environments for Advanced Research testbed is used. The testbed allows for the controlled movement of a pinned condition attached to an otherwise free cantilever beam. In previous studies, the testbed has been used with the generalized eigenvalue solver in a frequency-based model assimilation approach to infer the most likely position of the pinned condition (i.e., state of the structure). This work reports the effectivity of th

关键词： Finite element method

A QoS-Aware Software Defined Mobility Architecture for Named Data Networking

学校读者我要写书评

暂无评论

A QoS-Aware Software Defined Mobility Architecture for Named...

2022 IEEE GLOBECOM Workshops, GC Wkshps 2022

作者： Ali, Jehad Adnan, Muhammad Gadekallu, Thippa Reddy Jhaveri, Rutvij H. Roh, Byeong-Hee Sejong University Department of Computer Science and Engineering Seoul Korea Republic of Kookmin University Department of Electronics Engineering Seoul Korea Republic of Vellore Institute of Technology School of Information Technology and Engineering Vellore India Pandit Deendayal Energy University Department of Electronics Engineering Gandhinagar India Ajou University Department of AI Convergence Network Suwon Korea Republic of

ISBN: (纸本)9781665459754

Named Data Networking (NDN) is a future architecture for the internet and is an inventive model to provide content-based services. The unique properties of NDN, like naming schemes for network packets and approaches for in-network caching, makes it more reasonable for content distribution. However, extreme changes are required in the existing network infrastructure in order to implement NDN on it. One of the primary research dimensions for this new internet architecture is mobility. Although some research work has been done to address mobility concerns, there are still mobility issues like maintaining the connection establishment by frequently changing the position of users, loss of interest and data packets, interest and data packet transmission and re-transmission without any correspondence delay. One of the reasonable approaches is the Software Defined Networking (SDN) technique, which is used to realize mobility management for NDN. The core concept of SDN is the separation of the network control plane from the data plane. The data plane is capable of data forwarding, whereas the control plane is responsible for the entire network's control. The network control plane is decoupled from the data plane, resulting in a more programmable environment and the ability for external software to define a network's behavior. The SDN's main features, such as programmability, flexibility, and centralized control, make it a simple and scalable network. To cope with the issue of NDN mobility, we propose the design of a software-defined mobility architecture for NDN in which the SDN paradigm is extended to support mobility handling in the NDN architecture. Moreover, the intra-domain and Inter-domain handover handling scenarios in the software-defined mobility architecture for NDN have been described. © 2022 IEEE.

关键词： Internet of things

Federated Learning with Partial Model Personalization

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Pillutla, Krishna Malik, Kshitiz Mohamed, Abdelrahman Rabbat, Michael Sanjabi, Maziar Xiao, Lin Paul G. Allen School of Computer Science & Engineering University of Washington United States Meta AI United States

We consider two federated learning algorithms for training partially personalized models, where the shared and personal parameters are updated either simultaneously or alternately on the devices. Both algorithms have been proposed in the literature, but their convergence properties are not fully understood, especially for the alternating variant. We provide convergence analyses of both algorithms in the general nonconvex setting with partial participation and delineate the regime where one dominates the other. Our experiments on real-world image, text, and speech datasets demonstrate that (a) partial personalization can obtain most of the benefits of full model personalization with a small fraction of personal parameters, and, (b) the alternating update algorithm outperforms the simultaneous update algorithm by a small but consistent margin. Copyright © 2022, The Authors. All rights reserved.

关键词： Learning algorithms

Intelligent IoT Driven Ecosystem for Real-Time air and Noise Pollution Detection

学校读者我要写书评

暂无评论

Intelligent IoT Driven Ecosystem for Real-Time Air and Noise...

Cognitive Computing in engineering, Communications, sciences and Biomedical Health Informatics (IC3ECSBHI), International Conference on

作者： Shiv Narain Gupta Mohd Firoz Warsi Nitin Kumar Chauhan Rakhi Bhardwaj Mohd Jawed Khan Ashish Choudhary Department of Electronics and Communication Engineering Greater Noida Institute of Technology (Engineering Institute) Greater Noida Uttar Pradesh India Department of Computer Science & Engineering(AIDS) Greater Noida Institute of Technology (Engineering Institute) Greater Noida Uttar Pradesh India Department of Electronics and Communication Engineering Indore Institute of Science and Technology Indore Madhya Pradesh India Department of Artificial Intelligence and Machine Learning College of Smart Computing CORE University Roorkee Haridwar Uttarakhand India Department Computer Science & Engineering (AI&ML) Greater Noida Institute of Technology (Engineering Institute) Greater Noida Uttar Pradesh India

ISBN: (数字)9798331518523

ISBN: (纸本)9798331518530

The high speed of urbanisation and industrialisation necessitate keeping track of environmental parameters like air quality and noise pollution for the sustainability of ecology as well as the health of the environment and public health. By facilitating real-time data collection, analysis, and management, the Internet of Things (IoT) addresses challenges in an efficient way. The present work introduces an IoT based smart system for the real-time detection of both air and noise pollution. The developed system employs strategically placed sensor nodes to gather data on air quality parameters (e.g., particulate matter (PM2. 5, PM10), carbon monoxide (CO), nitrogen dioxide (NO 2), sulfur dioxide (SO 2), and ozone (O 3), along with sound levels measured in décibels (dB). These nodes have inexpensive sensors and wireless communication modules, sending the data to a central server for data processing.

关键词： Wireless communication Pollution Noise air quality Real-time systems Internet of Things Sulfur Sustainable development Monitoring Wireless fidelity

Towards Reliable and Empathetic Depression-Diagnosis-Oriented Chats

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Lan, Kunyao Ming, Cong Yao, Binwei Chen, Lu Wu, Mengyue X-LANCE Lab. Dept. of Computer Science and Engineering China MoE Key Lab of Artificial Intelligence AI Institute China Shanghai Jiao Tong University China University of Wisconsin Madison United States

Chatbots can serve as a viable tool for preliminary depression diagnosis via interactive conversations with potential patients. Nevertheless, the blend of task-oriented and chitchat in diagnosis-related dialogues necessitates professional expertise and empathy. Such unique requirements challenge traditional dialogue frameworks geared towards single optimization goals. To address this, we propose an innovative ontology definition and generation framework tailored explicitly for depression diagnosis dialogues, combining the reliability of task-oriented conversations with the appeal of empathy-related chit-chat. We further apply the framework to D4, the only existing public dialogue dataset on depression diagnosis-oriented chats. Exhaustive experimental results indicate significant improvements in task completion and emotional support generation in depression diagnosis, fostering a more comprehensive approach to task-oriented chat dialogue system development and its applications in digital mental health. Copyright © 2024, The Authors. All rights reserved.

关键词： Diagnosis

Influence of Contextual Information on Bengali-English Forward and Backward Transliteration Using Binary Coding

学校读者我要写书评

暂无评论

Influence of Contextual Information on Bengali-English Forwa...

Optimization Techniques for Learning (ICOTL), International Conference on

作者： Md. Fahad Amrita Das Tipu Ashis Kumar Mandal Basabi Chakraborty Department of Computer Science and Engineering Hajee Mohammad Danesh Science and Technology University Dinajpur Bangladesh Department of Computer Science University of Saskatchewan Saskatoon Canada Research and Regional Cooperation Division Iwate Prefectural University Takizawa Japan School of Computing Madanapalle Institute of Technology and Science Madanapalle India

Transliteration of words across different scripts carries importance for both social media platforms and messaging apps. It preserves the phonetic meaning of words across various writing systems. This study evaluates the influence of contextual information within words on the transliteration of Bengali and English language pair. It utilizes the Dakshina and NEWS 2018 datasets and binary encoding for feature representation. With the random forest classifier, the results demonstrate that using neighboring transliteration units as context yields the best performance on both datasets, obtaining the highest accuracy of 81.04%. These findings highlight the vital role of contextual information within words in enhancing the performance of transliteration.

关键词：