Correcting the hebbian mistake: Toward a fully error-driven hippocampus

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The hippocampus is crucial for episodic memory formation.
• Hebbian learning models, while common, are suboptimal for memory retrieval.
• Previous models used error-driven learning in CA1 and entorhinal cortex (EC).

Purpose of the Study:

• To propose a new computational model, Theremin, extending error-driven learning to hippocampal area CA3.
• To investigate Theremin's effectiveness in improving memory capacity and learning speed.
• To provide a biologically plausible mechanism for error-driven learning in the hippocampus.

Main Methods:

• Developed the Theremin (Total Hippocampal ERror MINimization) model.
• Modeled CA3 responding to EC input with a temporal difference via dentate gyrus (DG).
• Compared Theremin's performance against a Hebbian-based model.

Main Results:

• Theremin demonstrated significantly increased memory capacity compared to the Hebbian model.
• The model showed enhanced learning speed.
• Error-driven learning in CA3, guided by DG, reduced pattern interference.

Conclusions:

• Theremin offers a more effective computational framework for hippocampal memory than Hebbian learning.
• The model provides a biologically plausible explanation for error-driven learning in the hippocampus.
• Theremin generates testable predictions for future neuroscientific research.