How The Hippocampus Learns 

This piece, by Onno Berkan, was published on 11/05/24. The original text, by Zheng et al., was published by PLOS Computational Biology on 10/11/22.

This Randall O’Reilly study proposes a new model that predicts hippocampal behavior better than Hebbian learning. 

The hippocampus is a crucial brain region for forming everyday memories. Traditionally, scientists believed that the hippocampus learned using "Hebbian learning"– the idea that neurons that fire together strengthen their connections. However, this study introduces a new model called "Theremin" that uses "error-driven learning" instead.

The critical difference is that error-driven learning actively corrects mistakes and stops learning once it achieves successful memory recall. This is more efficient than Hebbian learning, which continues strengthening connections even when it's no longer necessary.

The model focuses on a region called CA3, the heart of the hippocampal memory system. It compares two states: an initial response to input and a slightly delayed, more refined response guided by another region called the dentate gyrus. The difference between these states creates an error signal that drives learning.

The researchers found that this new approach significantly improved memory capacity and learning speed. It was particularly good at keeping similar memories distinct and preventing them from interfering with each other.

While the model maintains many traditional ideas about how the hippocampus works, it suggests that error-driven learning might be a fundamental principle of memory formation. The idea here is tha,t as models of the hippocampus get better at approximating hippocampal behavior, the technologies and algorithms they implement get closer and closer to reality. What’s learned from this model could help explain how our relatively small hippocampus can store so many distinct memories throughout our lives.

The model also explains why testing yourself helps you remember better than reviewing information—testing creates opportunities for error correction and learning. This makes the model valuable for understanding basic brain function and explaining real-world learning phenomena.

Want to submit a piece? Or trying to write a piece and struggling? Check out the guides here!

Thank you for reading. Reminder: Byte Sized is open to everyone! Feel free to submit your piece. Please read the guides first though.

All submissions to berkan@usc.edu with the header “Byte Sized Submission” in Word Doc format please. Thank you!