ZUNA1.1: Flexible EEG Foundation Model for Denoising and Super-resolution

Christopher Warner, Jonas Mago, JR Huml, Beren Millidge· July 31, 2026 View original

Key takeaways

  • ZUNA1.1 is a flexible EEG foundation model for reconstruction.
  • It handles variable lengths, channels, and temporal intervals.
  • The model outperforms standard EEG denoising methods.
  • ZUNA1.1 is open-source, promoting accessibility and collaboration.

Who benefits

HealthcareResearch & AcademiaMedical DevicesNeuroscienceAI Development

Summary

ZUNA1.1 is a new 380M-parameter diffusion autoencoder designed as a flexible EEG foundation model for signal reconstruction. It can handle variable-length sequences, arbitrary channel numbers and locations, and reconstructs specific temporal intervals, outperforming standard EEG denoising methods while being open-source.

Researchers have released ZUNA1.1, an enhanced version of their EEG foundation model, now offering significantly greater flexibility for signal reconstruction tasks like denoising and super-resolution. This 380-million-parameter diffusion autoencoder is designed to process electroencephalography (EEG) data with unprecedented adaptability. ZUNA1.1 can reconstruct EEG sequences of variable lengths, up to 30 seconds, and accommodates an arbitrary number of EEG channels at any scalp location. Crucially, it also allows for the reconstruction of arbitrary temporal intervals within channels, not just entire channels. The model maintains performance on par with its predecessor, ZUNA1, while vastly expanding its utility. It consistently outperforms widely used standard EEG denoising and reconstruction methods, such as spherical spline interpolation found in the MNE package. ZUNA1.1 is released open-source under the Apache 2.0 license, promoting wider adoption and development.

Why it matters

This flexible and high-performing EEG foundation model can significantly advance neuroscience research, clinical diagnostics, and brain-computer interface development by providing superior signal processing capabilities.

How to implement this in your domain

  1. 1Download and integrate the open-source ZUNA1.1 model into existing EEG data processing pipelines for denoising and reconstruction.
  2. 2Experiment with ZUNA1.1's capabilities for handling variable-length EEG sequences and arbitrary channel configurations in your research.
  3. 3Compare ZUNA1.1's performance against current EEG denoising methods used in your lab or clinic.
  4. 4Explore using ZUNA1.1 for super-resolution tasks to enhance the quality of low-resolution EEG recordings.

Original post by Christopher Warner, Jonas Mago, JR Huml, Beren Millidge

"arXiv:2607.27308v1 Announce Type: new Abstract: We introduce ZUNA1.1, a 380M-parameter diffusion autoencoder for flexible EEG signal reconstruction. ZUNA1.1 is capable of reconstructing variable length sequences of up to 30s, with an arbitrary number of EEG channels at arbitrary…"

View on X

Originally posted by Christopher Warner, Jonas Mago, JR Huml, Beren Millidge on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses