Representational drift reflects ongoing balancing of stochastic changes by Hebbian learning.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Sensory processing involves neural representations that can change over time, a phenomenon termed representational drift.
• The underlying mechanisms driving representational drift, even in stable conditions, are not fully understood.

Purpose of the Study:

• To investigate the contributions of activity-dependent (Hebbian-like) and activity-independent (stochastic) synaptic processes to representational drift.
• To explore how these processes interact to shape neural dynamics in the auditory cortex.

Main Methods:

• Analysis of signal and noise correlations in neuron pairs using chronic calcium imaging data from the mouse auditory cortex.
• Development and testing of simple linear network models to simulate observed neural dynamics.

Main Results:

• Signal correlations were found to predict future noise correlations, indicating that stimulus-evoked coactivation strengthens effective connectivity.
• Linear network models successfully replicated temporal dependencies between signal and noise correlations when incorporating both Hebbian-like plasticity and stochastic synaptic changes.

Conclusions:

• Representational drift results from a dynamic interplay between Hebbian-like plasticity and stochastic synaptic modifications.
• Continuous sensory input-driven Hebbian-like plasticity acts as a counterbalance to stochastic changes, preserving network function and preventing degradation.