Predictive sequence learning in the hippocampal formation.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• The hippocampus is crucial for sequence encoding and memory.
• Understanding the precise computational mechanisms of the hippocampus remains a challenge.

Purpose of the Study:

• To develop a predictive autoencoder model of the hippocampal trisynaptic and monosynaptic circuits.
• To investigate how the hippocampus predicts sequences and signals prediction errors.

Main Methods:

• Developed a self-supervised recurrent neural network model of the hippocampus (entorhinal cortex, CA3, CA1).
• Trained CA3 to predict its next input.
• Analyzed spike coupling in simultaneously recorded neurons to confirm CA3's predictive capabilities.
• Modeled CA1 neurons as prediction error signalers.

Main Results:

• The model successfully replicated rapid place cell emergence and slow fading.
• Observed sequence replay and phase precession originating from CA3.
• Demonstrated biologically plausible learning through error-encoding neurons.
• CA1 neurons effectively signaled prediction errors.

Conclusions:

• The hippocampus utilizes a predictive coding mechanism for sequence learning.
• This computational motif, involving prediction and error signaling, may extend to other brain circuits like the thalamocortical system.