MANAS-2
Constrained reconstruction for EEG Foundation Models
About Model
Mannas AI releases MANAS-2, our brand new and state-of-the-art EEG Foundation Model, improving on its predecessor, MANAS-1. Built with 92M parameters and a new raw/band hybrid architecture, and driven to structured latent organisation through constrained reconstruction, MANAS-2 provides extremely information-rich embeddings from EEG signals. Across a wide vareity of downstream tasks, MANAS-2 significantly outperforms its competitors. PS - MANAS-2 is now an accepted paper at the NeurIPS BrainBodyFM Workshop 2026! Read the paper here: https://arxiv.org/abs/2609.13717.
MANAS-2
Metadata
GNU General Public License, version 3
Mannas AI
Feature Extraction
PyTorch
Open
Sector Agnostic
26/10/26 12:02:46
354.22 MB
config.json ( 55.75 KB )
To preview this file, you need to be a registered user. Please complete the registration process to gain access and continue viewing the content.
Activity Overview
0
0 
354.22 MB 
7
Tags
- AI
- Foundation Models
- EEG
- Biosignals
- MANAS
License Control
GNU General Public License, version 3
Version Control
Version 1(354.22 MB)
- Arvasu Yogesh Kulkarni·5 day(s) ago
- chevron_right
MANAS-2
config.json
LICENSE
model.py
model.safetensors
positions.json
README.md
requirements.txt
More Models from Mannas AI (Intellihealth World Pvt. Ltd.)
- 0
- 0
- 354.22 MB
8
- 0
- 0
- 1.50 GB
- 5
- 0
- 5
- 1.65 GB
- 142
© 2026 - Copyright AIKosh. All rights reserved.