DyBaNeM: Bayesian Framework for Episodic Memory Modelling

Abstract

Episodic Memory (EM) plays a central role in cognitive architectures and artificial agents that need ability to remember and recall the past. Most of the currently implemented EM systems resemble logs of events stored in a database with indexes. However, these systems usually lack abilities of human memory like hierarchical organization of episodes, reconstructive memory retrieval and encoding of episodes with respect to previously learnt repeated schemata. Here, we present a new general framework for EN modelling, DyBaNeM, that has these abilities. From psychological point of view, it builds on the Fuzzy-Trace Theory (FTT). On computational side, it uses formalism of Dynamic Bayesian Networks (DBNs). We describe abstract encoding, storage and retrieval algorithms, two models we implemented within DyBaNeM, and evaluation of the models' performance using a dataset resembling a log of human activity.


Back to Table of Contents