Representational drift and learning-induced stabilization in the piriform cortex.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Olfactory System

Background:

• The brain forms internal representations of stimuli via neural activity patterns.
• Representational drift (RD), the change in neural representations over time, is experimentally observed but poorly understood.
• Mechanisms driving RD in neural systems require further investigation.

Purpose of the Study:

• To investigate the mechanisms underlying representational drift (RD) in the piriform cortex.
• To model RD using a neural network incorporating synaptic weight dynamics at different timescales.
• To explain the relationship between RD, learning, and stimulus presentation frequency.

Main Methods:

• Developed a realistic neural network model of the piriform cortex.
• Incorporated two synaptic weight dynamics: slow multiplicative fluctuations (days) and spike-timing-dependent plasticity (STDP) (seconds).
• Analyzed the model's ability to reproduce experimental findings on RD and learning effects.

Main Results:

• Slow synaptic fluctuations quantitatively reproduced empirical RD effects.
• Fast STDP learning dynamics reduced representational dimensionality, suppressing RD.
• Model explained slower drift for learned odors and frequency-dependent drift rates, aligning with data.

Conclusions:

• Slow synaptic fluctuations are a key mechanism driving representational drift.
• Fast plasticity, like STDP, counteracts RD by stabilizing neural representations.
• The model provides a framework for understanding neural representation dynamics and learning in the brain.