Latent Computing by Biological Neural Networks: A Dynamical Systems Framework

Dinc, F., Blanco-Pozo, M., Klindt, D., Acosta, F., Jiang, Y., Ebrahimi, S., Shai, A., Tanaka, H., Yuan, P., Schnitzer, M. J., & Miolane, N.

We propose a dynamical systems framework centered on latent processing units to explain how neural circuits perform robust computations despite representational drift, identifying five key attributes that link low-dimensional computation, coding redundancy, linear decodability, and stability across time and scale.