Competitive intrinsically motivated agents for resource allocation

• Main Author: Zhao, Guopeng
• Other Authors: Bi Guoan
• Format: Theses and Dissertations
• Language: English
• Published: 2013
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; DRNTU::Engineering::Computer science and engineering::Information systems::Models and principles
• Online Access: https://hdl.handle.net/10356/53648
• Institution: Nanyang Technological University

Resource Allocation maps user requirements to specific resources, and it plays an important role in enabling a ubiquitous, convenient, on-demand network access to a shared pool of computing resources. Due to the dynamic nature of resource availability and the uncertainty of runtime environment, a successful allocation service should be flexible and agile. In particular, the service should be able to answer the three questions: (1) how can the service evolve with environmental changes that cannot be anticipated during design time; (2) how can the service make reliable and comprehensive decisions from incomplete information; and (3) how can the service systematically integrate and choose from a diversity of capabilities to achieve its objectives. To answer the above questions, Intrinsic Motivated Agent is viewed as a tailored approach to service adaptiveness, for an agent is considered as an autonomous problem solver, which can provide a “brain” for the resource allocation service to exhibit flexible and agile behaviors, and the agent with intrinsic motivations can further bring the service a unified sense of “self-willing” so that diverse behaviors can be integrated systematically. To this end, the thesis proposes a Competitive Intrinsically Motivated Agent by incorporating the psychological Self-Determination Theory (SDT) into agent modeling. The proposed approach of Competitive Intrinsically Motivated Agents for Resource Allocation enables an allocation service with enhanced proactiveness and improves performance in terms of the degree of adaptiveness that the service achieves. The proposed motivated agent is driven by the pursuit of three intrinsic needs, namely competence, relatedness and autonomy. In particular, agent’s behaviors in the pursuit of needs address each of the three questions individually: The need of competence drives agents to adapt to the dynamically changing environment, by a constant pursuit of good performance and new skills; The need of relatedness enables agents to provide reliable solutions from unevenly distributed information, by advocating collaborations with other agents to collect and synthesize information; The need of autonomy steers agents in systematically integrating various strategies and behaviors by encouraging them to make their own decisions proactively. The contributions of this thesis are summarized as follows. First, in light of SDT the research proposes a formal model of competitive intrinsically motivated agent, which takes account of agent’s intrinsic needs. Second, in order to bridge the gap between the proposed agent theoretical model and practical implementation, a goal-oriented approach is adopted to represent the proposed agent by systematically integrating agent’s psychological attributes, such as motivation and needs, with agent’s design artifacts such as goals, tasks and behaviors. Third, for resource allocation, a competitive intrinsically motivated agent is realized by incorporating a set of machine learning techniques, followed by discussions on the achieved agent adaptiveness. The resultant agent-mediated service benefits both service providers and service consumers through a robust, agile and adaptive resource sharing. Finally, two case studies are conducted to explore the practical use of the proposed agents for resource allocation in real-life applications. One is for load balancing in a Large-Scale Collaborative Virtual Environment, and the other is related to Cloud resource management for enterprise users. The evaluation results show that competitive agents with intrinsic motivations can exhibit adaptiveness in a great extend in terms of learning, performance, collaboration and decision making. Therefore, the success of the two case studies suggests that the proposed agent is not only promising but also practical.