Statistics of cortical representational drift can enable robust readout.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Representational drift, the reconfiguration of neural population codes, occurs across brain areas over days to weeks.
• The brain's ability to maintain information fidelity despite this drift is a key question.

Purpose of the Study:

• To investigate the impact of statistical properties of representational drift on information processing.
• To determine if downstream brain regions can adapt to neural drift and maintain information fidelity.

Main Methods:

• Analyzing the statistical properties of neural population drift, distinguishing between gradual changes and abrupt jumps.
• Developing and simulating an adaptive readout mechanism exploiting drift statistics.
• Examining experimental data from posterior parietal cortex and visual cortex for drift characteristics.

Main Results:

• Representational drift statistics are not typically Gaussian random walks.
• Sudden, heavy-tailed jumps in neural tuning are observed, rather than solely gradual changes.
• An adaptive readout mechanism can effectively use these heavy-tailed statistics for stable information extraction.

Conclusions:

• The statistical nature of representational drift significantly influences how downstream brain regions process information.
• Heavy-tailed drift statistics, characterized by sudden jumps, offer an advantage for adaptive readout mechanisms.
• Further research is needed to explore the existence and physiological basis of adaptive decoding in the brain.