Use this URL to cite or link to this record in EThOS:
Title: Recursive delegation and a trust-measuring algorithm
Author: Afanador, Juan
ISNI:       0000 0004 7972 5492
Awarding Body: University of Aberdeen
Current Institution: University of Aberdeen
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
This work is concerned with delegation as a recursive process, and the measuring of trust with respect to the shortcomings of the Eigentrust algorithm. These two subjects are related but inde-pendent, and hereby addressed as such. In formal and quotidian contexts, delegation signifies a series of interrelations associated with the relay of a task and the responsibility of its execution. As a model of interaction, delegation is invoked in applications as dissimilar as sensor networks and the outsourcing of banking services. Notwithstanding this relevance, current approaches overlook the possibility of further delegation from within active delegation events, and how critical this recursive aspect of delegation may be in known and new applications. Here, we use a game-theoretic framework to model recursive delegation. We use non-cooperative and coalitional approaches, which are shown to supersede existing delegation algo-rithms adjusted to recursive domains. We provide empirical evidence of such performance over varied network topologies, ranging from directed trees to random networks. Trust is often identified with a metric of reliability, enabling agents to operate in environ-ments fraught with structural uncertainty. The Eigentrust algorithm is representative of a class of algorithms which compute trust based on the connectivity of the graph underpinning every multi-agent system. In this sense, Eigentrust is founded upon firm theoretical guarantees, but also constrained by them, as per the Perron-Frobenious Theorem and its reliance on strongly-connected components. We generalise this approach by inscribing the trust-measuring problem within the Max-Plus Algebra. We do this by differentiating between having no information about an agent's trustwor-thiness, and being disconnected from it. The resulting algorithm preserves Eigentrust's theoretical guarantees, while displaying an improved performance in general multi-agent systems which op-erate on graphs of arbitrary topology.
Supervisor: Oren, Nir ; Baptista, Murilo Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: Algorithms ; Recursive functions ; Machine learning