Estimating the amount of computation done by a brain using population neural activity

Area of Science:

• Computational Neuroscience
• Systems Biology
• Dynamical Systems Theory

Background:

• Dynamical systems are often described as computational, yet quantifying this computation remains challenging.
• Existing research has focused on system dynamics, leaving the underlying computations largely unquantified.

Purpose of the Study:

• To introduce a task-independent framework for estimating computation in observed systems from time-series data.
• To define computation based on complexity and fidelity of statistical reconstructions.

Main Methods:

• Developed a framework to statistically reconstruct system dynamics from time-series data.
• Quantified computation by measuring reconstruction complexity and fidelity.
• Validated the framework using Lorenz dynamics and cellular automata.
• Applied the framework to neural recordings from *Caenorhabditis elegans* and mouse cortex.

Main Results:

• The framework successfully distinguished computation levels across different dynamical regimes and computational classes.
• Neural dynamics in *C. elegans* showed varying computation amounts related to movement.
• Mouse cortical activity during a visual task reflected task difficulty, with ambiguous inputs yielding higher computation.

Conclusions:

• The study presents a powerful, task-independent framework for quantifying computation in dynamical systems.
• The findings highlight the utility of this framework for analyzing neural computation in both simple and complex organisms.