Feasible closed-loop stimulation for neural latent dynamics control

In biophysically grounded spiking neural networks, using only ~20% of stimulation channels achieved ~85% of peak closed-loop latent control performance.

• Why it matters: Latent dynamics are rarely perturbed causally in vivo, and it is unclear what stimulation dimensionality and resource allocation are needed for closed-loop control.
• What they did: A high-throughput simulation survey tested channel allocation and stimulation/recording splits in spiking neural networks, evaluating tracking error and evoked behavior accuracy.
• The result: Splitting channels roughly equally between stimulation and recording optimized control, yielding >60% tracking error reduction and >90% evoked behavior accuracy on average, robust across parameter variations.

AI summary of the abstract · may contain errors · not medical advice · read the paper.