{"ID":23475084,"CreatedAt":"2026-09-18T01:09:05.407443952Z","UpdatedAt":"2026-09-20T18:11:56.143995915Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2609.19842","arxiv_id":"2609.19842","title":"Beyond Flattened Tokens: Structure-Preserving EEG Decoding with Reusable TriDim Blocks","abstract":"Effective EEG decoding requires representations that preserve organization among channels, local waveform dynamics, and long-range temporal context. Existing EEG architectures often capture these structures using separate specialized modules or collapse them into a single token sequence, making it difficult to maintain their distinct roles and coordinate their interactions throughout the backbone. We propose TriDim, a reusable block that preserves the representation shape and keeps three EEG axes explicit: channel, sample position within each patch, and patch position across the recording. These axes correspond to spatial, short-term temporal, and long-term temporal information, respectively. Each TriDim block applies feed-forward transformations along individual axes and cross-axis attention to coordinate information exchange among them. By stacking TriDim blocks with a multi-level tri-axis readout, we construct TriDimEEG, a standalone EEG decoder. Under strict cross-subject evaluation on eight datasets spanning clinical diagnosis, sleep staging, motor imagery, and emotion recognition, TriDimEEG achieves the best overall performance among fifteen evaluated models, with a 4.3% relative improvement in average accuracy over the second-best model. Replacing Transformer blocks in three EEG foundation models with TriDim blocks yields an average relative improvement of 7.4% in downstream accuracy while reducing parameter counts by 17.0% to 47.3%. These results establish TriDim as an effective and reusable building block and TriDimEEG as a strong standalone EEG decoder. Code and parameters of TriDimEEG are available at https://github.com/ncclab-sustech/TriDim_model.","short_abstract":"Effective EEG decoding requires representations that preserve organization among channels, local waveform dynamics, and long-range temporal context. Existing EEG architectures often capture these structures using separate specialized modules or collapse them into a single token sequence, making it difficult to maintain...","url_abs":"https://arxiv.org/abs/2609.19842","url_pdf":"https://arxiv.org/pdf/2609.19842v1","authors":"[\"Shiyue Su\",\"Song Wang\",\"Zekai Zhan\",\"Junjie Zeng\",\"Ziling Lu\",\"Zongsheng Li\",\"Xinyuan Ye\",\"Zhiyuan Ma\",\"Xinke Shen\",\"Quanying Liu\"]","published":"2026-09-17T07:52:27Z","proceeding":"cs.LG","tasks":"[\"cs.LG\"]","methods":"[\"Transformer\",\"Generative Adversarial Network\"]","has_code":false,"code_links":[{"ID":639790,"CreatedAt":"2026-09-18T01:09:05.407443952Z","UpdatedAt":"2026-09-18T01:09:05.407443952Z","DeletedAt":null,"paper_id":23475084,"paper_url":"https://arxiv.org/abs/2609.19842","paper_title":"Beyond Flattened Tokens: Structure-Preserving EEG Decoding with Reusable TriDim Blocks","repo_url":"https://github.com/ncclab-sustech/TriDim_model","is_official":false,"mentioned_in_paper":false,"mentioned_in_github":true,"github_stars":0}]}