Factorial Hidden Markov Models
Machine Learning - Special issue on learning with probabilistic representations
An Introduction to Variational Methods for Graphical Models
Machine Learning
A Bayesian Computer Vision System for Modeling Human Interactions
IEEE Transactions on Pattern Analysis and Machine Intelligence
| Hi-index | 0.00 |
A formalization of multi-agent systems (MAS) as hidden Markov models (HMM) is proposed and investigated from a view point of interaction among agents and environments. Conventional formalizations of agents as HMM do not take changes of environments in account, so that it is hard to analyze behaviors of agents that act in dynamic environments. The proposed formalization enables HMM to handles changes of environment and interaction among agents via environment directly inside of state-transitions. I first investigate HMM that represents changes of the environment in the same state-transitions of agent itself. Then I derive a structured model in which environment, agent, and another agent are treated as separated state-transitions and coupled with each other. For this model, in order to reduce the number of parameters, I introduce "symmetricity" among agents. Moreover, I discuss relation between reducing dependency in transitions and assumption of cooperative behaviors in MAS.