Associative Memory of Structured Knowledge

A long standing challenge in biological and artificial intelligence is to understand how new knowledge can be constructed from known building blocks in a way that is amenable for computation by neuronal circuits. While previous work has focused primarily on working memory tasks of structured data, here we focus on the task of storage and recall of structured knowledge in long term memory. Specifically, we ask how a Hopfield type network can store and retrieve episodic memories where each episode is a set of associations between events. We model each knowledge structure as a set of binary relations between events and cues (cues may represent e.g., temporal order, spatial location, role in semantic structure). We use a binarized version of holographic reduced representation (HRR) to map such structures to fixed length vectors. We then train a recurrent network to store these vectors as fixed points. By a combination of signal-to-noise analysis and numerical simulations we demonstrate that our model allows for an efficient storage and recall of these knowledge structures in a way that allows the full retrieval of the memorized structure and their building blocks from partial retrieving cues. Our work contributes to the understanding of neural computations of structured knowledge.