Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
CSBrain: A Cross-scale Spatiotemporal Brain Foundation Model for EEG Decoding
Authors: Yuchen Zhou, Jiamin Wu, Zichen Ren, Zhouheng Yao, Weiheng Lu, Kunyu Peng, Qihao Zheng, Chunfeng Song, Wanli Ouyang, Chao Gou
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments across 11 representative EEG tasks and 16 datasets demonstrate that CSBrain consistently outperforms both task-specific models and strong foundation baselines. |
| Researcher Affiliation | Collaboration | 1Shanghai Artificial Intelligence Laboratory 2Sun Yat-sen University 3The Chinese University of Hong Kong 4Karlsruher Institut fur Technologie |
| Pseudocode | No | The paper describes the model architecture and components (CST, SSA) using text and mathematical equations, but does not present any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code and model are available at https://github.com/yuchen2199/CSBrain. |
| Open Datasets | Yes | To comprehensively evaluate the generalizability of our model, we conduct experiments on 11 representative BCI tasks spanning 16 publicly available EEG datasets, as summarized in Table 1. |
| Dataset Splits | Yes | In all experiments, we strictly follow the training, validation, and test splits to ensure fair and consistent evaluation. To ensure fair and consistent evaluation, we adopt a subject-independent split: subject 1-5 for training, 6-7 for validation, and 8-9 for testing. |
| Hardware Specification | Yes | We use a batch size of 128 and train for 40 epochs on 4 NVIDIA A100 GPUs, with a total pre-training time of approximately 101 hours. |
| Software Dependencies | Yes | Pre-training is conducted using Python 3.11.11 and Py Torch 2.5.1 with CUDA 12.4. |
| Experiment Setup | Yes | Pre-training is conducted using Python 3.11.11 and Py Torch 2.5.1 with CUDA 12.4. The model is trained with the Adam W optimizer, a learning rate of 5e-4, weight decay of 5e-2, and cosine annealing learning rate scheduling. We use a batch size of 128 and train for 40 epochs on 4 NVIDIA A100 GPUs, with a total pre-training time of approximately 101 hours. |