Active inference relies on state-space models to describe the environments that agents sample with their actions. These actions lead to state changes intended to minimize future surprise. We show that surprise minimization relying on Bayesian inference can be achieved by filtering of the sufficient statistic time series of exponential family input distributions, and we propose the hierarchical Gaussian filter (HGF) as an appropriate, efficient, and scalable tool for active inference agents to achieve this.
Hierarchical gaussian filtering of sufficient statistic time series for active inference / Mathys, C.; Weber, L.. - 1326:(2020), pp. 52-58. (Intervento presentato al convegno 1st International Workshop on Active Inference, IWAI 2020 held in conjunction with the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, ECML-PKDD 2020 tenutosi a Ghent, Belgium nel September 14, 2020) [10.1007/978-3-030-64919-7_7].
Hierarchical gaussian filtering of sufficient statistic time series for active inference
Mathys C.
;
2020-01-01
Abstract
Active inference relies on state-space models to describe the environments that agents sample with their actions. These actions lead to state changes intended to minimize future surprise. We show that surprise minimization relying on Bayesian inference can be achieved by filtering of the sufficient statistic time series of exponential family input distributions, and we propose the hierarchical Gaussian filter (HGF) as an appropriate, efficient, and scalable tool for active inference agents to achieve this.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
