Review
Biochemical Research Methods
Wenze Ding, Kenta Nakai, Haipeng Gong
Summary: This review explores the recent advances in deep learning-based protein design methods and compares them with conventional knowledge-based approaches through notable cases. The development of deep learning in structure-based and sequence-based protein design is described, as well as the applications of deep reinforcement learning in protein design. Future perspectives on design goals, challenges, and opportunities are thoroughly discussed.
BRIEFINGS IN BIOINFORMATICS
(2022)
Article
Computer Science, Artificial Intelligence
Yu-Cen Lin, Chiao-Ting Chen, Chuan-Yun Sang, Szu-Hao Huang
Summary: The growing popularity of quantitative trading has attracted attention from traders and investment firms. A computational method for evaluating risk factors and returns is crucial for algorithmic trading strategies. This study proposes a multiagent deep reinforcement learning framework for effective portfolio management and achieves good results.
APPLIED SOFT COMPUTING
(2022)
Article
Computer Science, Artificial Intelligence
Menglong Lin, Tao Chen, Honghui Chen, Bangbang Ren, Mengmeng Zhang
Summary: The design of System of Systems (SoS) has been a topic of great concern, especially in military applications. This paper proposes a deep reinforcement learning approach called DRL-SoSDP to address the challenges of SoS architecting. By combining artificial intelligence techniques and actor-critic algorithms, DRL-SoSDP achieves superior results in solution quality and computation time, even in large scale cases.
ADVANCED ENGINEERING INFORMATICS
(2023)
Article
Computer Science, Artificial Intelligence
Yuanzheng Li, Shangyang He, Yang Li, Yang Shi, Zhigang Zeng
Summary: This article proposes a federated multiagent deep reinforcement learning algorithm for energy management in multimicrogrids. The federated learning mechanism is introduced to ensure data privacy and security. Experimental results demonstrate the effectiveness of the proposed method.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Engineering, Electrical & Electronic
Ye Yuan, Jiawan Zhang
Summary: In this paper, an unsupervised video summarization approach via reinforcement learning with shot-level semantics is proposed. The approach utilizes an encoder-decoder model to extract convolutional feature matrix from the video using a convolutional neural network as an encoder. A bidirectional LSTM is then used as a decoder to obtain probability weights for selecting keyframes to preserve spatio-temporal dependence. A shot-level semantic reward function is designed to reduce the influence of user subjectivity in generating more representative summarization results. The approach outperforms others and achieves satisfactory results according to evaluation on four classical datasets.
IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS FOR VIDEO TECHNOLOGY
(2023)
Article
Automation & Control Systems
Rui Yan, Xiaoming Duan, Zongying Shi, Yisheng Zhong, Jason R. Marden, Francesco Bullo
Summary: This article introduces a metric based on game-theoretic solution concept for the evaluation, ranking, and computation of policies in multiagent learning. The method can handle dynamical behaviors in multiagent reinforcement learning and is also compatible with single-agent reinforcement learning.
IEEE TRANSACTIONS ON AUTOMATIC CONTROL
(2022)
Article
Computer Science, Artificial Intelligence
Kaifang Wan, Dingwei Wu, Bo Li, Xiaoguang Gao, Zijian Hu, Daqing Chen
Summary: This paper proposes a new ME-MADDPG algorithm to improve the efficiency and adaptability of multiagent motion planning methods by introducing a mixed experience strategy. Experimental results demonstrate that the proposed algorithm significantly enhances convergence speed and effectiveness in training compared to traditional MADDPG, showing better performance in complex dynamic environments.
INTERNATIONAL JOURNAL OF INTELLIGENT SYSTEMS
(2022)
Article
Automation & Control Systems
Tianpei Yang, Weixun Wang, Jianye Hao, Matthew E. Taylor, Yong Liu, Xiaotian Hao, Yujing Hu, Yingfeng Chen, Changjie Fan, Chunxu Ren, Ye Huang, Jiangcheng Zhu, Yang Gao
Summary: The proposed Action Semantics Network (ASN) explicitly represents the influence of different actions on other agents in a multiagent system, improving the performance of deep reinforcement learning algorithms. Experimental results show significant enhancements compared to other network architectures, indicating promising applications in complex scenarios.
AUTONOMOUS AGENTS AND MULTI-AGENT SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Yulin Zhang, William Macke, Jiaxun Cui, Sharon Hornstein, Daniel Urieli, Peter Stone
Summary: This article explores the potential of autonomous vehicles in improving traffic congestion. By training multiagent driving policies, they can adapt to different traffic flows, AV penetration, and road geometries. The article also introduces a new cell transmission model suitable for simulating congestion in traffic networks, which can be used to study congestion reduction strategies.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Construction & Building Technology
Ling-chu Su, Xiqiang Wu, Xiaoning Zhang, Xinyan Huang
Summary: This study introduced a smart framework for fire-engineering Performance-Based Design (PBD) using Artificial Intelligence (AI) to predict smoke motion and Available Safe Egress Time (ASET) in atrium fires. By training a CFD database with AI, the model can accurately predict smoke visibility profiles and ASET in a much shorter time compared to conventional methods, demonstrating the feasibility of using AI in fire safety design to reduce time and cost.
JOURNAL OF BUILDING ENGINEERING
(2021)
Article
Engineering, Industrial
Yifan Zhou, Bangcheng Li, Tian Ran Lin
Summary: This paper introduces a hierarchical coordinated reinforcement learning (HCRL) algorithm to optimize maintenance of large-scale multicomponent systems, with agent parameters and coordination relationships designed based on system characteristics, and a hierarchical structure established according to components' structural importance measures. The effectiveness of the algorithm is confirmed through validation on different systems, outperforming other methods including deep reinforcement learning.
RELIABILITY ENGINEERING & SYSTEM SAFETY
(2022)
Article
Engineering, Multidisciplinary
Jingda Wu, Zhiyu Huang, Zhongxu Hu, Chen Lv
Summary: Machine learning, with its limited intelligence, cannot replace humans in real-world applications due to its inability to handle various situations. It is important to introduce humans into the training loop of artificial intelligence (AI) to leverage their robustness and adaptability. In this study, a real-time human-guidance-based deep reinforcement learning (DRL) method is developed for autonomous driving policy training. With a novel control transfer mechanism, humans can intervene and correct the agent's actions in real time during training. The proposed method improves DRL efficiency and performance by fusing real-time human guidance actions into the training loop.
Article
Computer Science, Artificial Intelligence
Xinghu Yao, Chao Wen, Yuhui Wang, Xiaoyang Tan
Summary: This article proposes a method named SMIX(lambda) to learn a stable and generalizable centralized value function (CVF) through off-policy training. By using the lambda-return as a proxy for computing the temporal difference error, the modified QMIX network structure is adopted to train the model. Experiments demonstrate the significant advantages of the proposed SMIX(lambda) method in multiagent reinforcement learning.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Engineering, Aerospace
Shichun Huang, Tao Wang, Yong Tang, Yiwen Hu, Gu Xin, Dianle Zhou
Summary: Cooperative formation control of unmanned ground vehicles (UGVs) is a significant research hotspot in UGV applications, attracting increasing attention in both military and civil domains. Compared to traditional algorithms, reinforcement-learning-based algorithms equipped with artificial intelligence offer a lower complexity solution for real-time formation control. This paper proposes a distributed deep-reinforcement-learning-based algorithm to address the navigation, maintenance, and obstacle avoidance tasks of UGV formations. The algorithm's effectiveness and scalability are validated through formation simulation experiments of different scales.
Article
Computer Science, Information Systems
Junwei Zhang, Zhenghao Zhang, Shuai Han, Shuai Lue
Summary: This paper discusses the exploration issue in the PPO algorithm and proposes an exploration enhancement mechanism based on uncertainty estimation. By applying the exploration enhancement theory to the PPO algorithm, the IEM-PPO algorithm is proposed, and it is evaluated in experiments using the MuJoCo physical simulator. The results show that the IEM-PPO algorithm outperforms PPO in terms of sample efficiency and cumulative reward.
INFORMATION SCIENCES
(2022)