NeuroLens preprint reports more stable decoding across neural recording sessions

Researchers report in a preprint that NeuroLens improved decoding of decision-making and semantic task variables from chronic intracortical recordings in mice and humans, with more stable decoding over time than state-of-the-art baselines. The self-supervised model maps changing populations of recorded neurons into a shared representation and predicts future states within that representation, reducing sensitivity to transient recording variability. Pretraining across multiple days enabled generalization to future sessions and adaptation to unseen neural populations using few examples. The findings concern decoding recorded neural activity across sessions.