Thesis supervisor: László Gulyás
Location of studies (in Hungarian): ELTE, Faculty of Informatics Abbreviation of location of studies: ELTE
Description of the research topic:
Classic Multi-Agent Systems (MAS) are popular for complex tasks because they are robust, flexible and easily scalable compared to centralised solutions. These systems usually make decisions using a predefined set of rules or heuristics. Multi-Agent Reinforcement Learning (MARL) is a modern and increasingly popular subclass of MAS. MARL uses reinforcement learning in decision-making, enabling agents to self-organise and better adapt to dynamically changing environments. MARL systems are currently being applied in many industrial fields, such as smart traffic lights, network modelling and industrial robot control. With the advancement of artificial intelligence, interactions between different learning algorithms or agents are becoming increasingly common, making MARL an active and growing area of research.
Nowadays, one of the active areas of research is on increasing the reliability and stability of MARL systems. In contrast to classical approaches, MARL systems are often less robust. This instability is due to the larger state space of MARL and the heterogeneity of the agents, which may result in over-specialized agents. These agents make the whole system sensitive to external noise, which may come from faulty sensors or targeted attacks. Real-world conditions are often difficult to simulate accurately, which poses additional challenges for MARL systems. Currently, the solutions proposed in the literature focus on preparing models for these disturbances during training. In this way, the agents become more generalized, but in return their computational demand increases significantly and even the performance of the models may decrease.
In addition to the existing approaches in the literature, this research project aims at developing learning and modelling methods that can be used to design more universal and resilient agents with less computational effort. The project aims at studying and developing novel methods that balance between efficiency and robustness. Approaches to replace obsolete (aging) or faulty agents or agents under attack will be an important part of the research, as well as the investigation and development of system-level re-organization methods. In particular, facilitating rapid adaptation to common attack strategies to increase system resilience is of special interest.
Required language skills: angol Recommended language skills (in Hungarian): B1 Further requirements: MSc in one of the following fields: informatics, software engineering, computer science or related areas.
Fluent in English.
Number of students who can be accepted: 1
Deadline for application: 2024-05-31
2024. IV. 17. ODT ülés Az ODT következő ülésére 2024. június 14-én, pénteken 10.00 órakor kerül sor a Semmelweis Egyetem Szenátusi termében (Bp. Üllői út 26. I. emelet).
Login
