Please use this identifier to cite or link to this item: https://ptsldigital.ukm.my/jspui/handle/123456789/513370
Title: Affective decision-making engine for solving complex problems in computer games
Authors: Hafiz Mohd Sarim (P51821)
Supervisor: Abdul Razak Hamdan, Prof. Dr.
Keywords: Computer games -- Programming
Software engineering
Universiti Kebangsaan Malaysia -- Dissertations
Dissertations, Academic -- Malaysia
Issue Date: 6-Aug-2018
Description: Affective agents are autonomous intelligent software agents that are programmed to achieve a set of goals by deciding upon action choices within a particular problem environment through the emulation of psychological affect, i.e. valenced „feelings‟ elicited toward the current problem environment. The Affective Decision-Making Engine (ADME) used by affective agents is designed to allow it to cope with complex problem environments such as computer games. For this research, open, interactive, and context-sensitive environments found in games are models for complex problems. Current adaptive agents and classification agents face difficulty in coping with these environments: the problem presents neither a consistent dataset to optimize against, nor a stable population of elements to train behaviour. These factors further exacerbate their ability to fulfil all set goals. This research is motivated by the ability of living organisms to adapt and summarily classify completely new and unfamiliar situations similar to these complex problems. In organisms, emotions act as an internal cognitive mechanism to evaluate how good or bad any situation is to the organism. The organism is positively or negatively affected, and is driven to perform actions that mitigate this affect. To emulate this process of affect, ADME assigns an Affect Value (AV) for every feature possessed by observed elements or objects in the problem environment. The AV calculates the correlation coefficient of a feature against the change in each goal outcome value, in a table called an Affect Matrix (AM). Through the sum correlation, the AM summarizes each element’s overall affect to each goal, to discover significant goals to prioritize. Later, a homeostat balances the significant goals with the best action for mitigating the affect. By using correlation coefficients, this research has found that the ADME is able to discover features that affect the agent’s goal outcome values. Experimental results show that affective agents surpass state-transition and adaptive agents when converging towards best actions for significant goals in complex problems, though it requires greater memory usage. These research findings show that affective agents have great utility as single pass multi-goal reinforcement tools for unfamiliar and highly dynamic environments.,Ph.D.
Pages: 291
Call Number: QA76.76.C672H333 2018 3 tesis
Publisher: UKM, Bangi
Appears in Collections:Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Files in This Item:
File Description SizeFormat 
ukmvital_120613+SOURCE1+SOURCE1.1.PDF
  Restricted Access
3.49 MBAdobe PDFThumbnail
View/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.