Existence of feedback-controllable subspaces in neural population dynamics

Determine whether simultaneously recorded neural population dynamics contain particular subspaces that differ in their amenability to feedback control, and identify such subspaces when present to understand their role in behavior.

Background

The paper motivates a control-theoretic analysis of neural population activity by noting that cognition and behavior rely on feedback, yet tools to assess feedback controllability directly from observed neural dynamics have been lacking. The authors introduce Feedback Controllability Components Analysis (FCCA), a dimensionality reduction method designed to identify subspaces of linear dynamical systems that are maximally feedback controllable, contrasting it with PCA which they show aligns with feedforward controllability.

By establishing theoretical connections between non-normal dynamics and the divergence of PCA versus FCCA subspaces, and by applying FCCA to neural data, the work aims to address the question of whether there are specific subspaces within neural population activity that are especially amenable to feedback control. This explicitly stated unknown motivates the development and application of FCCA to empirical recordings.