检索结果-内蒙古大学图书馆

Divide and Repair: Using Options to Improve Performance of Imitation Learning Against Adversarial Demonstrations

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Dasgupta, Prithviraj Distributed Intelligent Systems Section Information Technology Division Naval Research Laboratory WashingtonDC United States

We consider the problem of learning to perform a task from demonstrations given by teachers or experts, when some of the experts' demonstrations might be adversarial and demonstrate an incorrect way to perform the task. We propose a novel technique that can identify parts of demonstrated trajectories that have not been significantly modified by the adversary and utilize them for learning, using temporally extended policies or options. We first define a trajectory divergence measure based on the spatial and temporal features of demonstrated trajectories to detect and discard parts of the trajectories that have been significantly modified by an adversarial expert, and could degrade the learner's performance, if used for learning, We then use an options-based algorithm that partitions trajectories and learns only from the parts of trajectories that have been determined as admissible. We provide theoretical results of our technique to show that repairing partial trajectories improves the sample efficiency of the demonstrations without degrading the learner's performance. We then evaluate the proposed algorithm for learning to play an Atari-like computer-based game called LunarLander in the presence of different types and degrees of adversarial attacks of demonstrated trajectories. Our experimental results show that our technique can identify adversarially modified parts of the demonstrated trajectories and successfully prevent the learning performance from degrading due to adversarial demonstrations. © 2023, CC BY.

关键词： Demonstrations

Reward Shaping for Improved Learning in Real-time Strategy Game Play

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Kliem, John Dasgupta, Prithviraj Distributed Intelligent Systems Section Information Technology Division U. S. Naval Research Laboratory WashingtonDC United States

We investigate the effect of reward shaping in improving the performance of reinforcement learning in the context of the real-time strategy, capture-the-flag game. The game is characterized by sparse rewards that are associated with infrequently occurring events such as grabbing or capturing the flag, or tagging the opposing player. We show that appropriately designed reward shaping functions applied to different game events can significantly improve the player's performance and training times of the player's learning algorithm. We have validated our reward shaping functions within a simulated environment for playing a marine capture-the-flag game between two players. Our experimental results demonstrate that reward shaping can be used as an effective means to understand the importance of different sub-tasks during game-play towards winning the game, to encode a secondary objective functions such as energy efficiency into a player's game-playing behavior, and, to improve learning generalizable policies that can perform well against different skill levels of the opponent. © 2023, CC BY.

关键词： Reinforcement learning

Synthetically Generating Human-like Data for Sequential Decision Making Tasks via Reward-Shaped Imitation Learning

学校读者我要写书评

暂无评论

arXiv 2023年

关键词： Computer games

intelligent troubleshooting of complex machinery 90

学校读者我要写书评

暂无评论

Intelligent troubleshooting of complex machinery

Proceedings of the 3rd international conference on Industrial and engineering applications of artificial intelligence and expert systems - Volume 1

作者： Philippe L. Davidson Mike Halasz Sieu Phan Suhayya Abu Hakima Information Technology Section Laboratory for Intelligent Systems Division of Electrical Engineering National Research Council Canada Ottawa Canada K1A 0R8

ISBN: (纸本)9780897913720

Proper maintenance and troubleshooting of complex mechanical equipment is a difficult task. A large amount of information, such as sensor data and previous repair actions, is available but infrequently used for interpretation by technical staff which is continually losing expertise due to turnover. Well structured knowledge-based systems can provide effective techniques for assisting in this task.A set of functional requirements for a troubleshooting advisory system is presented. Domain knowledge should be structured to make the system's operation more explicit to the knowledge engineer or end user. Reasoning must be carried out by navigating through a diagnostic network constructed from maintenance procedures and heuristics. The generic and modular approach used has the distinct advantage of making the task of creating diagnostic knowledge-bases easier. In addition, the large amount of information which is normally disseminated through many maintenance manuals can be easily accessible in a single computer environment, which in itself represents considerable savings in time for a mechanic. A generic system, called JETA (Jet Engine Troubleshooting Assistant) has been developed within this framework. In particular, JETA has been applied to troubleshoot the General Electric J85-CAN-15 jet engine that powers the CF-5 trainer fighters used by the Canadian Air Force.

关键词：

Large margin multi-modal multi-task feature extraction for image classification

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Luo, Yong Wen, Yonggang Tao, Dacheng Gui, Jie Xu, Chao School of Electronics Engineering and Computer Science Peking University Beijing China Division of Networks and Distributed Systems School of Computer Engineering Nanyang Technological University Singapore Centre for Quantum Computation & Intelligent Systems Faculty of Engineering & Information Technology University of Technology Sydney Sydney Australia Centre for Quantum Computation & Intelligent Systems Faculty of Engineering and Information Technology University of Technology Sydney 81 Broadway Street UltimoNSW2007 Australia Hefei Institute of Intelligent Machines Chinese Academy of Sciences Hefei230031 China Center for Research on Intelligent Perception and Computing National Laboratory of Pattern Recognition Institute of Automation Chinese Academy of Sciences Beijing100190 China

The features used in many image analysis-based applications are frequently of very high dimension. Feature extraction offers several advantages in high-dimensional cases, and many recent studies have used multi-task feature extraction approaches, which often outperform single-task feature extraction approaches. However, most of these methods are limited in that they only consider data represented by a single type of feature, even though features usually represent images from multiple modalities. We therefore propose a novel large margin multi-modal multi-task feature extraction (LM3FE) framework for handling multi-modal features for image classification. In particular, LM3FE simultaneously learns the feature extraction matrix for each modality and the modality combination coefficients. In this way, LM3FE not only handles correlated and noisy features, but also utilizes the complementarity of different modalities to further help reduce feature redundancy in each modality. The large margin principle employed also helps to extract strongly predictive features so that they are more suitable for prediction (e.g., classification). An alternating algorithm is developed for problem optimization and each sub-problem can be efficiently solved. Experiments on two challenging real-world image datasets demonstrate the effectiveness and superiority of the proposed method. Copyright © 2019, The Authors. All rights reserved.

关键词： Feature extraction

Detecting potentially harmful and protective suicide-related content on twitter: A machine learning approach

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Metzler, Hannah Baginski, Hubert Niederkrotenthaler, Thomas Garcia, David Section for Science of Complex Systems Center for Medical Statistics Informatics and Intelligent Systems Medical University of Vienna Austria Unit Suicide Research & Mental Health Promotion Department of Social and Preventive Medicine Center for Public Health Medical University of Vienna Austria Complexity Science Hub Vienna Austria Institute of Interactive Systems and Data Science Department of Computer Science and Biomedical Engineering Graz University of Technology Graz Austria Institute of Globally Distributed Open Research and Education Austria Institute of Information Systems Engineering Vienna University of Technology Vienna Austria

Background Research has repeatedly shown that exposure to suicide-related news media content is associated with suicide rates, with some content characteristics likely having harmful and others potentially protective effects. Although good evidence exists for a few selected characteristics, systematic and large-scale investigations are missing. Moreover, the growing importance of social media, particularly among young adults, calls for studies on the effects of content posted on these platforms. Objective This study applies natural language processing and machine learning methods to classify large quantities of social media data according to characteristics identified as potentially harmful or beneficial in media effects research on suicide and prevention. Methods We manually labeled 3202 English tweets using a novel annotation scheme that classifies suicide-related tweets into 12 categories. Based on these categories, we trained a benchmark of machine learning models for a multi-class and a binary classification task. As models, we included a majority classifier, an approach based on word frequency (TF-IDF with a linear SVM) and two state-of-the-art deep learning models (BERT, XLNet). The first task classified postings into six main content categories, which are particularly relevant for suicide prevention based on previous evidence. These included personal stories of either (1) suicidal ideation and attempts or (2) coping and recovery, calls for action intending to spread either (3) problem awareness or (4) prevention-related information, (5) reporting of suicide cases, and (6) other tweets irrelevant to these five categories. The second classification task was binary, and separated postings in the 11 categories that refer to actual suicide, from postings in the off-topic category, which use suicide-related terms in another meaning or context. Results In both tasks, the performance of the two deep learning models was very similar and better than the majority or th

关键词： Learning algorithms

AUTOMATED OPPOSING FORCES FOR TRAINING SIMULATIONS

学校读者我要写书评

暂无评论

NAVAL ENGINEERS JOURNAL 1995年第3期107卷 245-252页

作者： ANTON, PS MCADOW, KL SMITH, DB LEHNER, PE SHARP, MA Dr. Philip S. Anton:is currently the Terhnology Area Manager of Modeling and Reasoning in the MITRE Advanced Information Technologies Center. His research at MITRE has included behavioral models of military doctrine and command environmental modeling and reasoning for command forces and intelligent simulations of military entities. His other interests include neural networks artificial intelligence and intelligent agents. Dr. Anton received a U.S. degree in Computer Engineering from UCLA and an M.S. degree and a Ph.D in Computational Neuroscience from the University of California. Irvine After finishing nuclear power training Mr. McAdow served as an Engineering Department Division Officer on USS Mariano G. Vallejo (SSBN 658). He then worked as a civilian far the Naval Coastal Systems Center and then far Martin Marietta Aero and Naval Systems in the development of various submarine systems and unmanned undersea vehicles. David Smith:has been a member of the technical staff at MITRE s Advanced Information Technologies center since 1993. He has a B.S. degree in Computer Science and Mathematics from Albright College and an M.S. degree in Information and Computer Science from the Georgia Institute of Technology. Dr. Paul Lehner:is a part-time Principal Scientist in the MITRE Advanced Information Technologies Center. In addition. Dr. Lehner is an Associate Professor of Systems Engineering at George Mason University. His research at MITRE has included automated planning and scheduling adversarial planning and reasoning against uncertainty. Dr. Lehner received a B.S. degree in Psychology from Bethany College. He also holds M.S. degrees in Psychology and Mathematics and a Ph.D. in Mathematical Psychology from the University of Michigan. Prior to his current employment at MITRE and at George Mason University Dr. Lehner served as the Technical Director of the Decision Sciences Section at PAR Government Systems Corporation. After finishing nuclear power training Cdr. Sharp served as an Engineering Department Divisi

Providing realistic opposing forces is critical to the successful use of military training simulations. Unfortunately, a number of issues can make the manual control of realistic opposing forces difficult or unattainable, This paper explores these issues while discussing how Automatic Interactive Targets (AITs) can assist Training Exercise Controllers (TECs) in providing validated and realistic opposing forces in highly interactive situations. The features of the prototype Remote AIT Processing System (RAPS) are used to demonstrate how an AIT system can be designed to meet TEC requirements for automated entities, RAPS can provide remote control of AITs for existing or new systems while providing sufficient features to allow a TEC to appropriately select and control AITs for individual training exercises and crew proficiencies.

关键词：

Journal of information Science and Engineering: Editorial notice

学校读者我要写书评

暂无评论

Journal of information Science and Engineering 2005年第5期21卷 i-ii页

作者： Lee, Der-Tsai Amato, Nancy M. Chang, Shih-Fu Chen, Homer H. Chen, Tsuhan Hsu, Tsan-Sheng Hwang, Jenq-Neng Kuo, Sy-Yen Kuo, Tei-Wei Li, Chung-Sheng Tokuyama, Takeshi Wing, Jeannette Wu, Tzong-Chen Yen, John Computer Science at Texas A and M University United States Parasol Laboratory United States IEEE Transactions on Parallel and Distributed Systems Computing Research Association's Committee on the Status of Women in Computing Research CRA-W's Distributed Mentor Program United States Department of Electrical Engineering Columbia University United States Lab. United States College of Electrical Engineering and Computer Science National Taiwan University Taiwan IEEE Transactions on Circuits and Systems for Video Technology IEEE Department of Electrical and Computer Engineering Carnegie Mellon University Pittsburgh PA United States ACM ACM SIGACT IEEE Computer Society IICM Austria Research and Development of the Department Multimedia Signal Processing Technical Committee IEEE Signal Processing Society United States IEEE Transactions on Circuits and systems for Video Technology College of Electrical Engineering and Computer Science National Taiwan Ocean University Keelung Taiwan Department of Electrical Engineering National Taiwan University Taiwan Department of Computer Science and Information Engineering National Taiwan University Taipei Taiwan IEEE Technical Committee on Real-Time Systems Computer Science Division IBM T.J. Watson Research Center United States IBM Research Division Graduate School of Information Sciences Tohoku University Japan ACM IPSJ Mathematical Society of Japan Japan Department of Computer Science Computer Science Department Carnegie Mellon University United States National Academies of Science's Computer Science and Telecommunications Board United States Microsoft's Trustworthy Academic Advisory Board Intel Research Pittsburgh's Advisory Board United States Dartmouth's Institute for Security Technology Studies Advisory Committee Canada Sloan Research Fellowships Program Committee United States ACM Taiwan China Information Sciences and Technology Pennsylvania State University United States Laboratory for Intelligent Agents Penn State's School of Info

No abstract available

关键词：

POS1470-HPR KNOWING WHAT TO DO WITH THE DATA - A QUALITATIVE STUDY ON CHALLENGES OF USING SMARTPHONE-BASED ePROs IN RHEUMATOID ARTHRITIS

学校读者我要写书评

暂无评论

Annals of the Rheumatic Diseases 2022年 81卷 1081-1081页

作者： Y. Seidler T.S. Jørgensen P. Studenic H. Radner T. Nygaard N. Weibrecht N. Popper L.E. Kristensen T.C. Wilhelmer J. Rickmann E. Mosor V. Ritschl T. Stamm Medical University of Vienna Section for Outcomes Research Center for Medical Statistics Informatics and Intelligent Systems Vienna Austria Copenhagen University Hospital Frederiksberg and Bispebjerg Parker Institute Department of Rheumatology Copenhagen Denmark Medical University of Vienna Division of Rheumatology Department of Internal Medicine III Vienna Austria Karolinska Institutet Division of Rheumatology Department of Medicine (Solna) Stockholm Sweden Daman P/S Copenhagen Denmark dwh GmbH Vienna Austria Vienna University of Technology Institute for Information Systems Engineering Vienna Austria EULAR Young PARE Zurich Switzerland Austrian Rheumatism League None Maria Alm Austria The Parker Institute´s Patient Association Department of Rheumatology Copenhagen Denmark Medical University of Vienna Section for Outcomes Research Center for Medical Statistics Informatics and Intelligent Systems Wien Austria

Background Using patient-reported outcomes (PROs) has a long tradition in rheumatology, and PRO measurement is included in many composite indices evaluating disease progression and treatment response [1]. However, little is known about patients´ and health professionals´ (HPs) perceptions of using digitally collected PROs, the so-called ePROs, with a personal smartphone app. Objectives To identify main challenges in utilising ePROs for management and treatment of rheumatoid arthritis from patients' and HPs' perspectives. Methods We interviewed 25 people with rheumatoid arthritis (RA) and 17 HPs (nurses, rheumatologists, and physiotherapists) from Austria and Denmark. We used the RheumaBuddy app as a practical example to illustrate the digital data collection and the feedback that patients would get from entering their self-reported outcomes. Interviews were recorded and transcribed. We applied a qualitative thematic analysis to identify major themes using a procedure of rigorous coding. Analysis was done by two researchers, and conflicts were solved by consensus. Ethical approval was obtained in both countries. Results Three main themes emerged: 1) Being simple yet comprehensive; 2) Resources to interpret, use and act upon the collected data; and 3) Being reminded of the disease. Within the first theme, many valued the intuitiveness and simplicity of ePROs, especially when used as a monitoring tool in between clinical visits. HPs were concerned about not to overwhelm the patients with too many questions. On the other hand, the short ePROs asked in the app were not comprehensive enough to capture psychosocial and lifestyle aspects of the disease which were considered important both by patients and HPs. Within the second theme, patients and HPs expressed that ePROs could be the basis for shared decision making. Nevertheless, some patients had clearer ideas on making use of the feedback they could get from their self-reported data than the others. Participants from Den

关键词：