Geometry Linked to Untangling Efficiency Reveals Structure and Computation in Neural Populations

Area of Science:

• Neuroscience
• Computational Biology
• Data Science

Background:

• Cognitive tasks often involve distinguishing overlapping signals, a process crucial for behavior.
• Neural circuits transform sensory inputs into separable representations, but linking geometry to efficiency is challenging.

Purpose of the Study:

• Introduce a data-driven framework, Geometry Linked to Untangling Efficiency (GLUE) with manifold capacity theory.
• Link geometrical properties of neural activity to representational untangling and computational efficiency.

Main Methods:

• Developed the GLUE framework incorporating manifold capacity theory.
• Applied GLUE to seven diverse neuroscience datasets (multiple organisms, tasks, recording techniques).

Main Results:

• Task-relevant representations untangle across various domains, including cortical hierarchy, learning, and intrinsic dynamics.
• GLUE characterizes geometric mechanisms of representational untangling.
• Demonstrated GLUE's ability to explain efficient and robust computation.

Conclusions:

• GLUE provides a unified framework for analyzing information organization in neural data.
• The framework has broad applications in neuroscience, structural genomics, and interpretable AI.
• GLUE offers a powerful tool for understanding high-dimensional representations.