Masterarbeit
Q-Learning Trust Awareness for Decentralized Web Applications
Completion
2024/01
Research Area
Students
Davide Olmo Garofoli
Advisers
Description
In recent years, the web has undergone tremendous development and has become increasingly important. We make use of it every day for several hours, for everything from work to entertainment. This has caused the continuous increase in interest on the part of the public, as well as a parallel increase in investment and development. In recent years, the idea of the decentralized web is approached by different projects and institutions. Its redecentralization means the tendency to prefer a distributed approach, rather than relying on centralized data management. Mainly, it is motivated by avoiding content control and illegitimate data management of hosts which appear sometimes as walled gardens. Additionally, such a decentralization of data can improve privacy and data control for end-users.
However, one of the main challenges related to the idea of the decentralized web is trust awareness. In fact, in a decentralized environment, data must not only be retrieved from other hosts, but the problem arises of verifying its trustworthiness. It is necessary to consider the possibility that others may interact in a harmeful manner or spread false information via decentralized knowledge graphs or pods. At present, the common method of verifying the trustworthiness of data in a decentralized system is to have it validated by an external third party which verifies and certifies the data. On the contrary, in a distributed web, it is necessary to develop a method to verify the validity of the data alternative to this centralized solution. For this system to be truly decentralized, and not depend on central systems that provide trust awareness, each web application would have to have its own trust awareness. So, each web application should be able to retrieve data from the web and decide which data is trustworthy in real time. This can be difficult because of the heterogenicity of the data in the web and its dynamic nature. A viable solution is to use Q-learning, which is a model-free reinforcement learning algorithm that can coordinate the actions of various applications to achieve collective behaviour. One solution to this problem is to use the knowledge of all individuals to assign a trustworthiness value and create a trust model. This approach is called "Trust-based Multi-Agent Credit Assignment". In the context of Q-learning, there are several other potentially suitable trust awareness models.
The objective of this master thesis is to find an approach or combination of approaches to solve the previously mentioned problem in the context of trust awareness for web applications using Q-learning approaches. This particularly includes the state of the art regarding Q-learning and its trust awareness models. The demonstration of feasibility with an implementation demonstrator of the concept is part of this thesis as well as a suitable evaluation with exemplary use cases.
