| Cooperative multiagent probabilistic inference can be applied in areas such as building surveillance and complex system monitoring and diagnosis to reason about the states of the corresponding distributed uncertain domains. In the static cases---where the current state is history independent, multiply sectioned Bayesian networks (MSBNs) provide a solution when interactions within each agent are structured and those among agents are limited. However, in the dynamic cases---where the current state is affected by history, if each agent evolves separately using a single agent dynamic Bayesian network (DBN), the agents' inference will not guarantee exact posterior probabilities.; In this thesis, the theory of MSBNs is applied to dynamic multiagent probabilistic inference. The following three specific tasks are addressed: (1) the generation of MSBNs for empirically studying algorithms on MSBNs, (2) the construction of a dynamic multiagent probabilistic inference testbed that builds dynamic environments and simulates their states over time, and (3) the proposal of dynamic multiagent probabilistic inference methods.; For the first task, the generation of MSBNs is treated as a sequence of decisions. The space of each decision is constrained so that a minimum amount of backtracking occurs and the outcome of generation is always a legal MSBN. A suite of algorithms is presented and experimental results are shown.; For the second task, it is proposed that sequential digital circuits are simulated as dynamic test environments. Sequential digital circuits are easy to understand, and contain general issues related with uncertain reasoning. The testbed can build sequential digital circuits and simulate their states over time, where some of the logic gates and flip-flops can be set faulty with specified probabilities.; For the third task, it is proposed that the entire domain for a period of time is modeled into an MSBN and the state of the domain is reasoned about period by period exactly using the MSBN. The state of a set of entities is reasoned about by observing all variables that are relevant to them. A concept of the graphical observable Markov boundary of a set of variables is introduced to capture all such relevant observable variables. (Abstract shortened by UMI.)... |
