MERLIN

MERLIN (the Memory, RL, and Inference Network) is a model for unsupervised predictive memory in a goal-directed agent proposed by Greg Wayne, et al., in which memory formation is guided by a process of predictive modeling. The authors state animals execute goal-directed behaviors despite limited range and scope of their sensors and to cope, they explore environments and store memories maintaining estimates of important information not available. MERLIN used 3D virtual reality environments for which partial observability was severe and memories had to be maintained over long durations and demonstrated that it can solve canonical behavioral tasks in psychology and neurobiology without simplifying assumptions about the dimensionality of sensory input or the duration of experiences.