Abstract
Active Inference for Dynamic Bayesian Networks / 4008 Caner Komurlu
In supervised learning, many techniques focus on optimizing training phase to increase prediction performance. Active inference, a relatively novel paradigm, aims to decrease overall prediction error via selective collection of some labels based on relations among instances. In this research, we use dynamic Bayesian networks to model temporal systems and we apply active inference to dynamically choose variables for observation so as to improve prediction on unobserved variables
