Machine Learning in Secure Beliefs-Desires-Intentions Agents

Abstract

Agent-based modeling (ABM) is a modeling system comprised of autonomous decision-making entities, known as agents, interacting with each other and their environment. The Belief, Desires, and Intentions (BDI) model is used for behavioral modeling of agents to perform practical reasoning, deciding what actions to perform to reach a goal. Extensions of the BDI model have been applied for mapping context to actionable plans in complex, uncertain environments through supervised and reinforcement learning techniques, including Q-learning, first-order logical decision trees, Support Vector Machine (SVM), genetic algorithm, and others. All components of multiagent systems (MAS), BDI, and extensions to machine learning require a security architecture for preserving the confidentiality, integrity, and availability of the system and data. A secure design at the agent and system level is required for secure agent communications, identification and authentication, permissions, and system level security for the host platform. With the increasing complexity of distributed systems, MAS applications to security monitoring, control, and management are a growing research area. Systems of BDI agents can perform security objectives to maintain the environment’s security posture through security-oriented goals, plans, and objectives.