Masked Autoencoders Learn Perception-Relevant Representations from Resting State Neural Data

Aleksandr Kovalev, Antonio Lozano, Fabrizio Grani, Cristina Soto Sanchez, Leili Soo, Rocío López-Peco, Adrian Villamarin-Ortiz, Roberto Morollón Ruiz, María del Mar Ayuso Arroyave, Alfonso Rodil, Eduardo Fernández
Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026, PMLR 308:93-98, 2026.

Abstract

Clinical neuroprosthetics face a data bottleneck: labeled perception trials are scarce while hours of spontaneous neural activity are largely underutilized. Here, we test whether self-supervised learning can use these unlabeled datasets to improve perception decoding. We pretrained a masked autoencoder on 14.6 hours of spontaneous multiunit activity from an intracortical array in a blind participant’s V1. The model captured interpretable brain structure without supervision: V1’s spatial organization and perceptual state separation both emerged purely from its latent representations. To test these features, we used linear probing (logistic regression on the frozen latents) to measure performance on the data with stimulation. Perception decoding accuracy reached 84.1% on a general psychometric task. On the more difficult threshold-level task, accuracy reached 64.0%. This work shows that spontaneous cortical activity is not noise; it contains rich, task-relevant structure. Unsupervised pretraining on this data is a promising strategy to improve neural decoding.

Cite this Paper

BibTeX
@InProceedings{pmlr-v308-kovalev26a, title = {Masked Autoencoders Learn Perception-Relevant Representations from Resting State Neural Data}, author = {Kovalev, Aleksandr and Lozano, Antonio and Grani, Fabrizio and Soto Sanchez, Cristina and Soo, Leili and L\'{o}pez-Peco, Roc\'{i}o and Villamarin-Ortiz, Adrian and Moroll\'{o}n Ruiz, Roberto and Ayuso Arroyave, Mar\'{i}a del Mar and Rodil, Alfonso and Fern\'{a}ndez, Eduardo}, booktitle = {Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026}, pages = {93--98}, year = {2026}, editor = {Abbasi-Asl, Reza and Iqbal, Asim and Ito, Shinya and Arkhipov, Anton and Sanborn, Sophia}, volume = {308}, series = {Proceedings of Machine Learning Research}, month = {27 Jan}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v308/main/assets/kovalev26a/kovalev26a.pdf}, url = {https://proceedings.mlr.press/v308/kovalev26a.html}, abstract = {Clinical neuroprosthetics face a data bottleneck: labeled perception trials are scarce while hours of spontaneous neural activity are largely underutilized. Here, we test whether self-supervised learning can use these unlabeled datasets to improve perception decoding. We pretrained a masked autoencoder on 14.6 hours of spontaneous multiunit activity from an intracortical array in a blind participant’s V1. The model captured interpretable brain structure without supervision: V1’s spatial organization and perceptual state separation both emerged purely from its latent representations. To test these features, we used linear probing (logistic regression on the frozen latents) to measure performance on the data with stimulation. Perception decoding accuracy reached 84.1% on a general psychometric task. On the more difficult threshold-level task, accuracy reached 64.0%. This work shows that spontaneous cortical activity is not noise; it contains rich, task-relevant structure. Unsupervised pretraining on this data is a promising strategy to improve neural decoding.} }
Endnote
%0 Conference Paper %T Masked Autoencoders Learn Perception-Relevant Representations from Resting State Neural Data %A Aleksandr Kovalev %A Antonio Lozano %A Fabrizio Grani %A Cristina Soto Sanchez %A Leili Soo %A Rocío López-Peco %A Adrian Villamarin-Ortiz %A Roberto Morollón Ruiz %A María del Mar Ayuso Arroyave %A Alfonso Rodil %A Eduardo Fernández %B Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026 %C Proceedings of Machine Learning Research %D 2026 %E Reza Abbasi-Asl %E Asim Iqbal %E Shinya Ito %E Anton Arkhipov %E Sophia Sanborn %F pmlr-v308-kovalev26a %I PMLR %P 93--98 %U https://proceedings.mlr.press/v308/kovalev26a.html %V 308 %X Clinical neuroprosthetics face a data bottleneck: labeled perception trials are scarce while hours of spontaneous neural activity are largely underutilized. Here, we test whether self-supervised learning can use these unlabeled datasets to improve perception decoding. We pretrained a masked autoencoder on 14.6 hours of spontaneous multiunit activity from an intracortical array in a blind participant’s V1. The model captured interpretable brain structure without supervision: V1’s spatial organization and perceptual state separation both emerged purely from its latent representations. To test these features, we used linear probing (logistic regression on the frozen latents) to measure performance on the data with stimulation. Perception decoding accuracy reached 84.1% on a general psychometric task. On the more difficult threshold-level task, accuracy reached 64.0%. This work shows that spontaneous cortical activity is not noise; it contains rich, task-relevant structure. Unsupervised pretraining on this data is a promising strategy to improve neural decoding.
APA
Kovalev, A., Lozano, A., Grani, F., Soto Sanchez, C., Soo, L., López-Peco, R., Villamarin-Ortiz, A., Morollón Ruiz, R., Ayuso Arroyave, M.d.M., Rodil, A. & Fernández, E.. (2026). Masked Autoencoders Learn Perception-Relevant Representations from Resting State Neural Data. Proceedings of the First Workshop on NeuroAI Multimodal Intelligence @ AAAI 2026, in Proceedings of Machine Learning Research 308:93-98 Available from https://proceedings.mlr.press/v308/kovalev26a.html.

Related Material