Eleonora RussoOct 8, 2023
DurstewitzLab

Our Perspective on reconstructing computational system dynamics from neural data finally out in Nature Rev Neurosci!
https://www.nature.com/articles/s41583-023-00740-7

We survey generative models that can be trained on time series data to mimic the behavior of the underlying neural substrate.

Reconstructing computational system dynamics from neural data with recurrent neural networks - Nature Reviews Neuroscience

The prospects for applying dynamical systems theory in neuroscience are changing dramatically. In this Perspective, Durstewitz et al. discuss dynamical system reconstruction using recurrent neural networks to directly infer a formal surrogate from an experimentally probed system and consider its potential for revolutionizing neuroscience.

Nature
Oct 8, 2023 at 10:34amWeb
Show threadDurstewitzLabOct 8, 2023
With training algorithms incorporating control-theoretical ideas, RNNs can learn to behave like the underlying dynamical system they have been trained on, and the role of different latent states in dynamics can be dissected.
Show threadDurstewitzLabOct 8, 2023
Successful dynamical systems reconstructions should exhibit some generalization properties, like correctly inferring the long-term behavior of systems from finite-length time series, and generalizing at least to neighboring initial conditions.
Show threadDurstewitzLabOct 8, 2023
We also discuss various ways of linking model latent spaces to the biological substrate through specific configuration of decoder models. This adds another layer of mechanistic interpretability to such models.
Show threadDurstewitzLabOct 8, 2023
Forgot to include the share link: https://rdcu.be/dnILu