National and Kapodistrian University of Athens
Decision-making of autonomous agents in networks with adversaries
Abstract
dc:descriptionIn modern networks, agents are characterized by enhanced capabilities of independent decision making, resulting in improved network efficiency in terms of throughput maximization, energy consumption and scalability, but new challenges emerge. Agents may have disparate, or even conflicting interests and hence, “misbehavior” issues can arise. The term “misbehavior” refers to the case when agents act in a way that is misaligned with the one prescribed by network protocols. As a result, altruistic cooperation can notbe taken for granted. For this reason, mechanisms that can enforce agents’ compliance with a desired behavior that is beneficial for the network, have been proposed in literature. Such mechanisms aretrust/reputation models, which aim at incentivizing agents to act in a cooperative way for their own benefits. This thesis aims at investigating the interactions of autonomous network agents when they have individual interests, as well as the impact of cooperation stimulation mechanisms on their decision-making process. More specifically, we investigate when agents, with myopic selfish or malicious motives, opt to share information with each other undersome cooperation enforcing mechanism such as trust/reputation models or punishment strategies. We start by investigating the information sharing in the most basic network functionality, namely packet-forwarding. In this study the impact of trust and action monitoring mechanisms on the agents’ decision-making process is analyzed and conditions are derived which if satisfied, ensure agents are incentivized to opt to follow a strategy that is beneficial for the network. Then, we move on to study information sharing in a higher level, by considering a specific information diffusion model and study the problem of distributed parameter estimation in an adversarial setting, where malicious agents exist in the network having the option to disseminate intentionally falsified information. Novel algorithms that are robust to malicious information dissemination are developed to drive agents’ parameter estimation and decision-making processes. Finally, we make an abstraction and study the problem of information sharing in a more general setting, without assuming a specific information sharing model. We utilize the notion of conditional mutual information to quantify the value of shared information. Appropriate punishment strategies are devised that can provably sustain information sharing at equilibrium. To achieve the research goals of this thesis, methods from Stochastic Control Theory, Game Theory, Signal Processing and Learning algorithms are utilized.
Degree
thesis:*- Grantor dc:publisher
- National and Kapodistrian University of Athens
- Year dc:date
- 2019
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Ntemos, Konstantinos
- Ντέμος, Κωνσταντίνος
Subjects
dc:subject × 14- Λήψη αποφάσεων
- Δίκτυα αυτόνομων κόμβων
- Κακόβουλοι κόμβοι
- Μοντέλα εμπιστοσύνης
- Εκμάθηση
- Decision-making
- Networks of autonomous agents
- Malicious agents
- Trust/ reputation models
- Learning
- Φυσικές Επιστήμες
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Computer and Information Sciences
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/46658