A5100323308- Reinforcement Learning in Robotics
- Software System Performance and Reliability
- Software Engineering Research
- Adaptive Dynamic Programming Control
- Distributed Control Multi-Agent Systems
- Adversarial Robustness in Machine Learning
- Energy Efficiency and Management
- Game Theory and Applications
- Advanced Software Engineering Methodologies
- Software Reliability and Analysis Research
- Evolutionary Algorithms and Applications
- Economic Theory and Institutions
- Smart Grid Energy Management
- Scheduling and Optimization Algorithms
- Robot Manipulation and Learning
- Experimental Behavioral Economics Studies
Nanjing University
2022-2024
Institute of Software
2023
In the field of mixed-motive games, extensive multi-agent learning studies have explored balance between egoism (individual interest), utilitarianism (collective and egalitarianism (fairness). Traditional approaches often rely on manually designed reward functions, social norms, alliance/federation mechanisms to transition agents from individualistic behaviors toward cooperative strategies. However, these methods typically require all share private local information or mandatorily...
Abstract As a predominant design method for microsservices architecture (MSA), domain‐driven (DDD) utilizes series of standard patterns in both models and implementations to effectively support the architectural elements. However, an implementation may deviate from its original domain model that uses certain patterns. The deviation between is type drift , which needs be detected promptly. This paper proposes approach, namely DOMICO, check conformance implementation, by formalized defining...
Multi-task multi-agent reinforcement learning (MT-MARL) is capable of leveraging useful knowledge across multiple related tasks to improve performance on any single task. While recent studies have tentatively achieved this by independent policies a shared representation space, we pinpoint that further advancements can be realized explicitly characterizing agent interactions within these and identifying task relations for selective reuse. To end, article proposes Representing Interactions...
In cooperative multi-agent reinforcement learning, decentralized agents hold the promise of overcoming combinatorial explosion joint action space and enabling greater scalability. However, they are susceptible to a game-theoretic pathology called relative overgeneralization that shadows optimal action. Although recent value-decomposition algorithms guide by learning factored global value function, representational limitation inaccurate sampling actions during process make this problem still....
The deep reinforcement learning (DRL) algorithm works brilliantly on solving various complex control tasks. This phenomenal success can be partly attributed to DRL encouraging intelligent agents sufficiently explore the environment and collect diverse experiences during agent training process. Therefore, exploration plays a significant role in accessing an optimal policy for DRL. Despite recent making great progress continuous tasks, these tasks has remained insufficiently investigated. To...
The combination of deep reinforcement learning (DRL) with ensemble methods has been proved to be highly effective in addressing complex sequential decision-making problems. This success can primarily attributed the utilization multiple models, which enhances both robustness policy and accuracy value function estimation. However, there limited analysis empirical current RL thus far. Our new reveals that sample efficiency previous DRL algorithms may by sub-policies are not as diverse they...
Multi-agent systems are characterized by environmental uncertainty, varying policies of agents, and partial observability, which result in significant risks. In the context Multi-Agent Reinforcement Learning (MARL), learning coordinated decentralized that sensitive to risk is challenging. To formulate coordination requirements risk-sensitive MARL, we introduce Risk-sensitive Individual-Global-Max (RIGM) principle as a generalization (IGM) Distributional IGM (DIGM) principles. This requires...
Deep reinforcement learning (DRL) provides a new way to generate robot control policy. However, the process of training policy requires lengthy exploration, resulting in low sample efficiency (RL) real-world tasks. Both imitation (IL) and from demonstrations (LfD) improve by using expert demonstrations, but imperfect can mislead improvement. Offline Online lot offline data initialize policy, distribution shift easily lead performance degradation during online fine-tuning. To solve above...