检索结果-内蒙古大学图书馆

Consistent Action for Stable Training in Reinforcement Learning–based Gain Tuning of Linear Feedback Controller

Journal of Institute of Control, Robotics and Systems 2024年第9期30卷 965-972页

作者： Byun, Hyungjo ASRI Department of Electrical and Computer Engineering Seoul National University Korea Republic of

Controlling nonlinear systems with linear feedback controller after linearization is a widely used method. This paper proposes a new method to efficiently train a reinforcement learning agent to select the control gain of the linear feedback controller. The proposed method involves the agent selecting a consistent gain by choosing between a new and an old gain. Since the linear feedback controller can operate with this consistent gain, the agent does not need to select the optimal gain, thus simplifying the training process. Numerical simulations were performed by applying the proposed method to a longitudinal model of a nonlinear missile with a 3-loop autopilot. The proposed method enables training with a wide range of gains and enhances the reference tracking performance compared to classical gain scheduling. © ICROS 2024.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

How Does a Swarm Exhibit Emergent Behavior Through Synchronization?

引用

Journal of Institute of Control, Robotics and Systems 2024年第4期30卷 352-362页

作者： Shim, Hyungbo ASRI Electrical and Computer Engineering Department Seoul National University Korea Republic of

A swarm of individuals often exhibits behaviors that are not possible for each individual. This phenomenon is called emergence, and this paper mathematically demonstrates that new dynamics can arise in swarm behavior that cannot be explained by the dynamics of individuals. In particular, we argue that emergence occurs when heterogeneity is coupled with synchronization. These two concepts may seem conflicting, as heterogeneity is the tendency to differ whereas synchronization is the tendency toward sameness. However, we show that emergent behavior arises from the interplay between the two. © ICROS 2024.

关键词： Multi agent systems

来源：评论

学校读者我要写书评

暂无评论

Encrypting Controllers via Input-Output Representation by Ring-LWE based Cryptosystem

引用

Journal of Institute of Control, Robotics and Systems 2023年第1期29卷 1-6页

作者： Lee, Joowon ASRI Department of Electrical and Computer Engineering Seoul National University Korea Republic of

Since security issues have emerged significantly with sophisticated attacks against cyber-physical systems, the notion of encrypted control using homomorphic encryption was proposed, which enables controllers to compute over encrypted data without decryption on the cyber layer. There have been several results providing methods to encrypt various controllers and systems, however, encrypting dynamic controllers has remained challenging due to the limited number of repeated homomorphic operations available on ciphertexts. This paper proposes a method to encrypt linear dynamic controllers using Ring-LWE (Learning With Errors) based cryptosystem, by transforming them into input-output representation, where the controller output is expressed as a linear combination of previous inputs and outputs. This approach avoids recursive homomorphic operations at the encrypted controller, with the additional help of re-encryption, so that the controller is able to operate for infinite-time horizon without growing encryption errors. Furthermore, we utilize properties of Ring-LWE based cryptosystem in order to reduce the number of homomorphic operations inside the controller compared to previous results, without involving any additional algorithms such as key switching, besides homomorphic addition and multiplication. © ICROS 2023.

关键词： Networked control systems

来源：评论

学校读者我要写书评

暂无评论

Distributed Estimation of the Geometric Center of Multi-target by a Group of Agents With Bearing Measurements

引用

Journal of Institute of Control, Robotics and Systems 2023年第11期29卷 880-886页

作者： Hyeon, Soojeong ASRI Department of Electrical and Computer Engineering Seoul National University Korea Republic of

This paper presents a distributed estimation scheme for determining the geometric center of multiple targets by a group of agents in 2-dimensional space. These agents have the knowledge of bearing measurements to a part of the targets and communicate with each other. Specifically, we utilize a continuous-time dynamics with strong coupling among agents to process local information. This approach allows us to assign the equilibrium point of the dynamics describing collective behavior of the agents to the geometric center of the targets. The benefits of this approach include: (i) no initialization process is required and (ii) global information, such as the numbers of targets and agents, is not required. These properties make the proposed estimation method suitable for plug-and-play operation. © ICROS 2023.

关键词： Multi agent systems

来源：评论

学校读者我要写书评

暂无评论

CONFIDENCE-BASED FEATURE IMPUTATION FOR GRAPHS WITH PARTIALLY KNOWN FEATURES 11

CONFIDENCE-BASED FEATURE IMPUTATION FOR GRAPHS WITH PARTIALL...

引用

11th International Conference on Learning Representations, ICLR 2023

作者： Um, Daeho Park, Jiwoong Park, Seulki Choi, Jin Young Department of Electrical and Computer Engineering ASRI Seoul National University Korea Republic of

This paper investigates a missing feature imputation problem for graph learning *** methods have previously addressed learning tasks on graphs with missing ***, in cases of high rates of missing features, they were unable to avoid significant performance *** overcome this limitation, we introduce a novel concept of channel-wise confidence in a node feature, which is assigned to each imputed channel feature of a node for reflecting certainty of the *** then design pseudo-confidence using the channel-wise shortest path distance between a missing-feature node and its nearest known-feature node to replace unavailable true confidence in an actual learning *** on the pseudo-confidence, we propose a novel feature imputation scheme that performs channel-wise inter-node diffusion and node-wise inter-channel *** scheme can endure even at an exceedingly high missing rate (e.g., 99.5%) and it achieves state-of-the-art accuracy for both semi-supervised node classification and link prediction on various datasets containing a high rate of missing *** are available at https://***/daehoum1/pcfi. © 2023 11th International Conference on Learning Representations, ICLR 2023. All rights reserved.

关键词： Classification (of information)

来源：评论

学校读者我要写书评

暂无评论

Adversarial Environment Design via Regret-Guided Diffusion Models 38

Adversarial Environment Design via Regret-Guided Diffusion M...

引用

38th Conference on Neural Information Processing Systems, NeurIPS 2024

作者： Chung, Hojun Lee, Junseo Kim, Minsoo Kim, Dohyeong Oh, Songhwai Interdisciplinary Program in Artificial Intelligence ASRI Seoul National University Korea Republic of Department of Electrical and Computer Engineering ASRI Seoul National University Korea Republic of

Training agents that are robust to environmental changes remains a significant challenge in deep reinforcement learning (RL). Unsupervised environment design (UED) has recently emerged to address this issue by generating a set of training environments tailored to the agent's capabilities. While prior works demonstrate that UED has the potential to learn a robust policy, their performance is constrained by the capabilities of the environment generation. To this end, we propose a novel UED algorithm, adversarial environment design via regret-guided diffusion models (ADD). The proposed method guides the diffusion-based environment generator with the regret of the agent to produce environments that the agent finds challenging but conducive to further improvement. By exploiting the representation power of diffusion models, ADD can directly generate adversarial environments while maintaining the diversity of training environments, enabling the agent to effectively learn a robust policy. Our experimental results demonstrate that the proposed method successfully generates an instructive curriculum of environments, outperforming UED baselines in zero-shot generalization across novel, out-of-distribution environments. Project page: https://***/projects/ADD. © 2024 Neural information processing systems foundation. All rights reserved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Exponential Synchronization of Heterogeneous Multi-Agent Systems via Linear-Signum-Type Diffusive Coupling: Blended Dynamics Approach 23

Exponential Synchronization of Heterogeneous Multi-Agent Sys...

引用

23rd International Conference on Control, Automation and Systems, ICCAS 2023

作者： Seong, Jeong Mo Lee, Donggil Shim, Hyungbo Seoul National University Asri Department of Electrical and Computer Engineering Seoul08826 Korea Republic of

ISBN: (纸本)9788993215274

This paper investigates the behavior of heterogeneous agents that interact through diffusive coupling, resulting in emergent blended dynamics that may not be observed in the dynamics of individual agents. In particular, we propose a linear-signum-type diffusive coupling to achieve exponential synchronization among heterogeneous agents. This finding offers a notable advantage over previous research, which primarily focused on linear-type diffusive coupling with remaining steady-state error in synchronization. We provide conditions of both linear and signum coupling strengths for achieving the exponential synchronization and analyze the associated convergence rate. Furthermore, simulation results validate the effectiveness of the proposed approach. © 2023 ICROS.

关键词： Dynamics

来源：评论

学校读者我要写书评

暂无评论

Attention-Based Randomized Ensemble Multi-Agent Q-Learning 23

Attention-Based Randomized Ensemble Multi-Agent Q-Learning

引用

23rd International Conference on Control, Automation and Systems, ICCAS 2023

作者： Park, Jeongho Kwon, Obin Oh, Songhwai Seoul National University Department of Electrical and Computer Engineering and Asri Seoul08826 Korea Republic of

ISBN: (纸本)9788993215274

Cooperative multi-agent scenarios are prevalent in real-world applications. Optimal coordination of agents requires appropriate task allocation, considering each task's complexity and each agent's capability. This becomes challenging under decentralization and partial observability, as agents must self-allocate tasks using limited state information. We introduce a novel multi-agent environment in which effective sub-task assignment is crucial for high-scoring performance. In addition, we propose a new multi-agent reinforcement learning framework named as attention-based randomized ensemble multi-agent Q-learning, or AREQ for short. This approach integrates a unique network structure using a multi-head attention mechanism, efficiently extracting task-related information from observations. AREQ also incorporates a randomized ensemble method, enhancing sample efficiency. We explore the impact of this attention-based structure and the random ensemble method through an ablation study and show AREQ's superiority compared to existing MARL methods within our proposed environment. © 2023 ICROS.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

CONTINUAL LEARNING IN THE PRESENCE OF SPURIOUS CORRELATIONS: ANALYSES AND A SIMPLE BASELINE 12

CONTINUAL LEARNING IN THE PRESENCE OF SPURIOUS CORRELATIONS:...

引用

12th International Conference on Learning Representations, ICLR 2024

作者： Lee, Donggyu Jung, Sangwon Moon, Taesup Department of Electrical and Computer Engineering Sungkyunkwan University Korea Republic of Department of Electrical and Computer Engineering Seoul National University Korea Republic of ASRI INMC IPAI AIIS Seoul National University Korea Republic of

Most continual learning (CL) algorithms have focused on tackling the stability-plasticity dilemma, that is, the challenge of preventing the forgetting of past tasks while learning new ones. However, we argue that they have overlooked the impact of knowledge transfer when the training dataset of a certain task is biased - namely, when the dataset contains some spurious correlations that can overly influence the prediction rule of a model. In that case, how would the dataset bias of a certain task affect the prediction rules of a CL model for future or past tasks? In this work, we carefully design systematic experiments using three benchmark datasets to answer the question from our empirical findings. Specifically, we first show through two-task CL experiments that standard CL methods, which are oblivious of the dataset bias, can transfer bias from one task to another, both forward and backward. Moreover, we find out this transfer is exacerbated depending on whether the CL methods focus on stability or plasticity. We then present that the bias is also transferred and even accumulates in longer task sequences. Finally, we offer a standardized experimental setup and a simple, yet strong plug-in baseline method, dubbed as group-class Balanced Greedy Sampling (BGS), which are utilized for the development of more advanced bias-aware CL methods. © 2024 12th International Conference on Learning Representations, ICLR 2024. All rights reserved.

关键词： Plasticity

来源：评论

学校读者我要写书评

暂无评论

Model Reference Gaussian Process Regression: Data-Driven State Feedback Controller for Strongly Controllable Systems 63

Model Reference Gaussian Process Regression: Data-Driven Sta...

引用

63rd IEEE Conference on Decision and Control, CDC 2024

作者： Kim, Hyuntae Seoul National University Asri Department of Electrical and Computer Engineering 1 Gwanak-ro Gwanak-gu Seoul08826 Korea Republic of

ISBN: (纸本)9798350316339

Data-driven control methods are gaining importance in control engineering, particularly for nonlinear systems where traditional models fall short. Many approaches rely on predefined libraries of functions, such as polynomials, to approximate system dynamics. These methods assume the selected functions can effectively represent the underlying system, but this assumption may not always hold, potentially impacting the robustness of control designs. The Model Reference Gaussian Process Regression (MR-GPR) controller was introduced to address these issues by using Gaussian Process Regression (GPR) with flexible kernels that adapt to observed data, eliminating the need for predefined function libraries. In this paper, we extend the MR-GPR controller to handle n-step controllable systems, allowing for more complex multi-step control tasks. We propose a state feedback control strategy based on realtime system data. Theoretical analysis confirms the stability of the closed-loop system under the MR-GPR controller, and numerical simulations validate its effectiveness. © 2024 IEEE.

关键词： State feedback

来源：评论

学校读者我要写书评

暂无评论

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

请选择保存的检索档案：

请选择收藏分类：

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

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

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：