Unsupervised Predictive Memory in a Goal-Directed Agent

TL;DR

This paper introduces MERLIN, a memory-augmented reinforcement learning model that effectively handles partial observability in complex 3D environments by using predictive modeling to guide memory formation.

Contribution

The paper presents MERLIN, a novel AI architecture that combines memory, reinforcement learning, and inference, enabling agents to solve complex, partially observable tasks without simplifying assumptions.

Findings

MERLIN outperforms traditional RL algorithms in partially observable 3D tasks.

Memory guided by predictive modeling improves long-term task performance.

The model operates effectively without assumptions on sensory input dimensionality.

Abstract

Animals execute goal-directed behaviours despite the limited range and scope of their sensors. To cope, they explore environments and store memories maintaining estimates of important information that is not presently available. Recently, progress has been made with artificial intelligence (AI) agents that learn to perform tasks from sensory input, even at a human level, by merging reinforcement learning (RL) algorithms with deep neural networks, and the excitement surrounding these results has led to the pursuit of related ideas as explanations of non-human animal learning. However, we demonstrate that contemporary RL algorithms struggle to solve simple tasks when enough information is concealed from the sensors of the agent, a property called "partial observability". An obvious requirement for handling partially observed tasks is access to extensive memory, but we show memory is not enough; it is critical that the right information be stored in the right format. We develop a model, the Memory, RL, and Inference Network (MERLIN), in which memory formation is guided by a process of predictive modeling. MERLIN facilitates the solution of tasks in 3D virtual reality environments for which partial observability is severe and memories must be maintained over long durations. Our model demonstrates a single learning agent architecture that can solve canonical behavioural tasks in psychology and neurobiology without strong simplifying assumptions about the dimensionality of sensory input or the duration of experiences.