Imprecision in Social Learning

Abstract

This thesis investigates the role of imprecision in social learning, which consists of two critical aspects: belief fusion with other agents and evidential updating based on direct evidence from the environment. A decentralised collective learning problem is investigated in which a population of agents attempts to learn the true state of the world and a novel parameterised fusion operator that allows varying levels of imprecision is proposed. This is used to explore the effect of fusion imprecision on learning performance in a series of agent-based simulations. In general, the results suggest that imprecise fusion operators are optimal when the frequency of fusion is high
relative to the frequency with which evidence is obtained from the environment. A parallel line of study explores the role of imprecision in evidential updating in a noisy environment. Through agent-based simulations, we demonstrate that the social learning model is robust to imprecise evidence. Our results also show that certain kinds of imprecise evidence can enhance the efficacy of the learning process in the presence of sensor errors. An integrated model is then proposed, combining the advantages of both the parameterised fusion operator and novel evidential updating strategies. We demonstrate that an integrated approach combining fusion and evidential imprecision can further enhance the robustness and accuracy of social learning processes. In addition, we compare various evidential updating methods for set-based belief and then propose a hybrid updating method to combine the strengths of different methods. We found that hybrid methods can enhance the accuracy and robustness of social learning significantly with more time required for the agents to reach consensus. These findings have significant implications for designing intelligent systems capable of social learning and decentralised decision-making.

Date of Award	9 Jan 2024
Original language	English
Awarding Institution	University of Bristol
Supervisor	Jonathan Lawry (Supervisor) & Michael Crosscombe (Supervisor)