Random compressed coding with neurons.

Area of Science:

• Computational neuroscience
• Neural coding theory

Background:

• Classical models assume simple neural tuning curves (e.g., bell-shaped).
• Complex neuronal responses, like those in grid cells, can yield highly accurate neural population codes.
• The necessity of fine-tuning for accurate neural coding remains an open question.

Purpose of the Study:

• To investigate whether highly accurate neural population codes necessitate finely tuned neuronal response properties.
• To explore the relationship between tuning curve irregularity and coding accuracy.

Main Methods:

• A computational model of a neural population with random, spatially extended, and irregular tuning curves was developed.
• The model analyzed the trade-off between local resolution enhancement and global error introduction due to irregularity.
• Information compression and coding accuracy were assessed under conditions balancing local and global errors.

Main Results:

• Irregular tuning curves enhance local coding resolution but can cause global errors.
• Optimal tuning curve smoothness balances local and global errors, leading to information compression.
• This compressed code achieves exponential accuracy, demonstrating "compressed efficient coding."
• Analysis of monkey motor cortex recordings supports the existence of such compressed efficient coding.

Conclusions:

• Highly accurate neural codes do not require fine-tuning of neuronal response properties.
• Efficient codes emerge robustly from irregularity and randomness in tuning curves.
• Compressed efficient coding provides a framework for understanding robust, high-accuracy neural representations.