REMI: Reconstructing Episodic Memory During Internally Driven Path Planning.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• Grid cells in the medial entorhinal cortex (MEC) form triangular spatial codes.
• Place cells in the hippocampus (HC) encode specific locations and environmental contexts.
• The parallel existence of these two spatial representations and their functional interplay remains incompletely understood.

Purpose of the Study:

• To propose a theoretical framework where grid and place cells cooperate to support cognitive planning.
• To investigate how place cells might recall goal locations using sensory and motivational cues, facilitating planning via grid cell representations.
• To explore the computational mechanisms underlying path integration and path forecasting in a neural network model.

Main Methods:

• Development of a single-layer recurrent neural network (RNN) model.
• Simultaneous modeling of the hippocampus-medial entorhinal cortex (HC-MEC) loop and a planning subnetwork.
• Extension of existing theoretical frameworks for grid-cell-based planning.

Main Results:

• Demonstration that local transition rules can enable long-distance path forecasting.
• Evidence that the planning network can sequentially update grid cell states toward a goal.
• Observation that intermediate grid activity can trigger place cell pattern completion, reconstructing planned paths.

Conclusions:

• The proposed model integrates HC-MEC function with planning capabilities.
• Recurrent mechanisms facilitate grid cell-based planning, with goal recall supported by the place system.
• The framework generates testable predictions regarding the neural basis of spatial planning and memory.