Continuous Bump Attractor Networks Require Explicit Error Coding for Gain Recalibration.

Area of Science:

• Computational neuroscience
• Neural networks
• Cognitive modeling

Background:

• Continuous bump attractor networks (CBANs) are models for representing continuous variables in brain functions like spatial navigation and working memory.
• CBANs integrate input to update representations but accumulate errors over time.
• Existing models lack mechanisms for recalibrating the integration gain, a process observed experimentally.

Purpose of the Study:

• To investigate the neural mechanisms behind integration gain recalibration in CBANs.
• To model experimental findings on gain recalibration in hippocampal place cells.
• To bridge the gap in understanding how CBANs achieve plasticity in their integration process.

Main Methods:

• Utilized a ring attractor network, a type of CBAN, to simulate experimental conditions.
• Analyzed the network dynamics required for gain recalibration.
• Developed a modified ring attractor network incorporating an error-rate code and Hebbian plasticity.

Main Results:

• Identified that gain recalibration, distinct from error correction, necessitates an explicit neural signal encoding representation error via a rate code.
• Demonstrated that ground-truth inputs fine-tune integration gain, a crucial factor for accurate representation updates.
• The proposed modified CBAN model successfully achieved integration gain recalibration.

Conclusions:

• Gain recalibration in CBANs requires specific neural mechanisms beyond simple error correction.
• An explicit error-rate code combined with Hebbian plasticity enables integration gain recalibration.
• This work provides a theoretical and computational framework for understanding neural plasticity in continuous variable representation.