AI Emulates Brain’s Memory Replay, Uncovering How We Learn

The hippocampus, a crucial brain region, plays a vital role in memory formation. This significance has been exemplified by notable cases, such as that of patient H.M., who experienced an inability to create new memories after the removal of substantial portions of his hippocampus.

Research conducted on rodents has shed light on the hippocampus’s involvement in spatial learning and navigation. One significant discovery in this domain is the identification of place cells, which are neurons that activate in specific locations.

Nicolas Diekmann explains, “These place cells are involved in a fascinating phenomenon called replay. As an animal moves, specific place cells fire in a sequence that corresponds to the animal’s path. Later, during rest or sleep, these same place cells can reactivate in the same order or even in reverse order.”

The observed sequences during replay not only mirror past behavior but can also be rearranged, adapting to changes in the environment or representing unvisited but observed locations.

“Our interest lies in understanding how the hippocampus efficiently produces various types of replay and their underlying purpose,” outlines Nicolas Diekmann.

To investigate this, the researchers constructed a computer model where artificial intelligence learns spatial information. Their ultimate goal is to study how quickly the AI agent can navigate and find an exit from specific spatial scenarios. The better the AI knows the environment, the faster it can achieve this goal.

Interestingly, the AI agent also learns by repeating neuronal sequences. However, these sequences are not randomly played back; instead, they are prioritized according to specific rules.

“Sequences are replayed stochastically based on their prioritization,” highlights Diekmann. Familiar sequences are given higher priority, and positions associated with rewards are replayed more frequently.

“Our model is biologically plausible, has manageable computational requirements, and learns faster compared to agents that randomly replay sequences,” summarizes Nicolas Diekmann. “This provides us with a deeper understanding of how the brain learns.”

Author: Neurologica

Leave a Reply