Statistical learning and representational drift: A dynamic substrate for memories

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Neurons exhibit continuous changes in tuning properties over days, known as representational drift.
• This drift occurs even when perception and behavior remain stable.
• Understanding how neuronal circuits maintain function amidst constant change is a key challenge.

Purpose of the Study:

• To review theoretical and experimental work on representational drift.
• To explore mechanisms maintaining stable function in neuronal circuits.
• To propose the role of statistical learning in stabilizing neural representations.

Main Methods:

• Review of existing theoretical and experimental literature.
• Analysis of neural dynamics from synaptic changes to population-level activity.
• Integration of concepts from statistical learning and neural coding.

Main Results:

• Representational drift arises from synaptic changes affecting individual neuron tuning.
• Population-level activity patterns can remain stable, preserving representational similarities.
• Statistical learning is proposed as crucial for maintaining representational stability under steady conditions.

Conclusions:

• Neuronal circuits maintain stable function through dynamic processes, not static codes.
• Statistical learning plays a vital role in preserving representational stability.
• This framework reconciles neural instability with perceptual stability, impacting understanding of learning, memory, and forgetting.