Name of the participant: Jannis Weil
Description of the IT research project: The importance of the Internet is greater today than ever before. Countless interconnected systems work non-stop to provide the services we need and process requests around the world. Many of the algorithms used for coordination and data exchange were designed manually. They use heuristics that provide sufficient, but not necessarily optimal, results. If systems perform sub-optimally, this can result in increased costs and poor user experience.
Therefore, there is a trend towards adaptive systems. They adjust to the requirements independently under a few specifications. Under laboratory conditions, they also achieve better results than previous heuristics in many use cases, such as video streaming and service placement. This project focuses on learning methods from the field of Reinforcement Learning. The systems considered interact with the network over several steps, make decisions and change its state. Although the Internet consists of many networked nodes, these methods are often only applied locally. Intelligent systems act independently of each other or only consider limited neighbors in the network. Alternatively, the entire network is controlled from a central location. However, this is not practical for large networks.
The overall goal of this project is to improve the adaptability of current reinforcement learning methods to networks. The focus is on the exchange of information between learning communication systems. The project investigates how they can exchange information autonomously and selectively throughout the network. This is expected to improve the coordination capabilities of the systems. In addition, methods are being researched to improve the generalizability of the procedures to different network topologies and scenarios.
Software Campus partners: Technische Universität Darmstadt, HUAWEI
Implementation period: 01.03.2022 – 29.02.2024
© project logo: Alisha Hirsch
