A non-Hebbian code for episodic memory.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Hebbian plasticity is the dominant model for memory formation.
• Existing models struggle to explain one-shot episodic memory, which rarely depends on both pre- and postsynaptic activity.

Purpose of the Study:

• To propose and validate a novel episodic memory model based on non-Hebbian plasticity.
• To demonstrate that simpler plasticity rules can effectively encode episodic memories.

Main Methods:

• Developed an episodic memory model utilizing a plasticity rule dependent solely on presynaptic activity.
• Leveraged high-dimensional neural activity with restricted transitions to store episodes as paths.
• Utilized an odor-tracking algorithm for decoding memory traces termed path vectors.

Main Results:

• The model naturally stores episodes as paths through complex state spaces, forming expressive path vectors.
• Path vectors serve as robust alternatives to Hebbian traces.
• Demonstrated support for one-shot sequential and associative recall, and policy learning.

Conclusions:

• Non-Hebbian plasticity is sufficient for flexible memory and learning.
• This model provides a framework for encoding episodes and policies as paths within a world model.
• Sheds light on potential hippocampal plasticity mechanisms.