Long-term stability of cortical population dynamics underlying consistent behavior

These results fall under the category 'the unreasonable effectiveness of linear methods'. Nonlinear methods are sometimes essential to progress: we love and use AEN, GAN, LFADS, and others. But sometimes a good understanding and use of linear methods is all that is needed. 2/n

— Sara A Solla (@SaraASolla) January 6, 2020

More from the tweet stream:

To me is still amazing that linear approaches suffice to align complex low-dimensional dynamics across days, months, even years! Not only the results are exciting, but we learned much in the process. 3/n

We are grateful to our Nature Neuroscience editor, who guided us and supported us through a long and difficult review process: three revisions over one full year. Some colleagues are very attached to the idea of a simple one-dimensional tuning curve to each neuron. 4/n

The problem resides with the belief that individual tuning curves can explain any and all observed neural activity, and with the refusal to accept the need for population analyses. So, to our most difficult and recalcitrant anonymous reviewer: I hope you learned something. 5/

The full abstract from the article linked in the top tweet:

Animals readily execute learned behaviors in a consistent manner over long periods of time, and yet no equally stable neural correlate has been demonstrated. How does the cortex achieve this stable control? Using the sensorimotor system as a model of cortical processing, we investigated the hypothesis that the dynamics of neural latent activity, which captures the dominant co-variation patterns within the neural population, must be preserved across time. We recorded from populations of neurons in premotor, primary motor and somatosensory cortices as monkeys performed a reaching task, for up to 2 years. Intriguingly, despite a steady turnover in the recorded neurons, the low-dimensional latent dynamics remained stable. The stability allowed reliable decoding of behavioral features for the entire timespan, while fixed decoders based directly on the recorded neural activity degraded substantially. We posit that stable latent cortical dynamics within the manifold are the fundamental building blocks underlying consistent behavioral execution.