--- license: cc-by-nd-4.0 language: - en library_name: pytorch tags: - eeg - biosignal - mamba - state-space-model - cross-attention - foundation-model - self-supervised - masked-modeling - lejepa - topology-invariant - neuroscience datasets: - TUEG - TUAB - APAVA - TDBrain - MoBI - SEED-V - Mumtaz2016 - MODMA metrics: - balanced_accuracy - roc_auc - pr_auc - r2 - pearson_r - cohen_kappa thumbnail: https://raw.githubusercontent.com/pulp-bio/BioFoundation/refs/heads/main/docs/model/logo/LuMamba_logo.png model-index: - name: LuMamba-Tiny (LeJEPA-reconstruction pre-training) results: - task: type: time-series-classification name: EEG Abnormality Detection dataset: type: TUAB name: TUH EEG Abnormal Corpus (TUAB) metrics: - type: balanced_accuracy value: 80.99 name: Balanced Accuracy (%) - type: roc_auc value: 0.883 name: AUROC - type: pr_auc value: 0.892 name: AUC-PR - task: type: time-series-classification name: Alzheimer's Disease Detection dataset: type: APAVA name: APAVA metrics: - type: roc_auc value: 0.955 name: AUROC - type: pr_auc value: 0.970 name: AUC-PR - task: type: time-series-classification name: Parkinson's Disease Detection dataset: type: TDBrain name: TDBrain metrics: - type: roc_auc value: 0.961 name: AUROC - type: pr_auc value: 0.960 name: AUC-PR - task: type: time-series-classification name: Major Depressive Disorder Detection dataset: type: Mumtaz2016 name: Mumtaz2016 metrics: - type: roc_auc value: 0.931 name: AUROC - type: pr_auc value: 0.952 name: AUC-PR - name: LuMamba-Tiny (Reconstruction-only pre-training) results: - task: type: time-series-classification name: EEG Slowing Event and Seizure Detection dataset: type: TUSL name: TUH EEG Slowing Corpus (TUSL) metrics: - type: roc_auc value: 0.708 name: AUROC - type: pr_auc value: 0.289 name: AUC-PR - task: type: time-series-classification name: EEG Artifact Detection dataset: type: TUAR name: TUH EEG Artifact Corpus (TUAR) metrics: - type: roc_auc value: 0.914 name: AUROC - type: pr_auc value: 0.510 name: AUC-PR - task: type: time-series-classification name: Gait Prediction Regression dataset: type: MoBI name: MoBI metrics: - type: r2 value: 0.116 name: R-squared - type: rmse value: 0.1482 name: Root Mean Squared Error - task: type: time-series-classification name: 5-class Emotion Detection dataset: type: SEED-V name: SEED-V metrics: - type: balanced_accuracy value: 35.0 name: Balanced Accuracy (%) - type: cohen_kappa value: 0.191 name: Cohen's Kappa - task: type: time-series-classification name: Major Depressive Disorder Detection dataset: type: MODMA name: MODMA metrics: - type: balanced_accuracy value: 59.5 name: Balanced Accuracy (%) - type: roc_auc value: 0.448 name: AUROC - type: pr_auc value: 0.420 name: AUC-PR ---
LuMamba Logo

LuMamba: Latent Unified Mamba for Electrode Topology-Invariant and Efficient EEG Modeling

Github License Paper

**LuMamba** (Latent Unified Mamba) is an **EEG foundation model** built on efficient **Mamba state-space learning**, capable of handling **heterogeneous channel topologies**. LuMamba addresses varying channel layouts with **LUNA channel unification**, projecting a given EEG channel layout to a **fixed latent topology**, and overcomes the quadratic complexity of transformers with **FEMBA**'s efficient **bidirectional Mamba encoder**. --- ## ๐Ÿ”’ License & Usage Policy (Weights) **Weights license:** The released model weights are licensed under **Creative Commons Attributionโ€“NoDerivatives 4.0 (CC BY-ND 4.0)**. This section summarizes the practical implications for users. *This is not legal advice; please read the full license text.* ### โœ… You may - **Use** and **redistribute** the **unmodified** LuMamba weights (including in commercial settings) **with proper attribution** to the LuMamba authors. - **Fine-tune / adapt** the weights **for your internal use** (research or production) **without redistributing** the modified weights. - **Publish your code, configs, logs, and papers** describing experiments with LuMamba (please cite the paper). ### ๐Ÿšซ You may not - **Share, host, or redistribute any modified weights** (including LoRA/adapter/delta checkpoints or pruned/quantized variants). Any parameter set that encodes an adaptation is considered a derivative and cannot be shared under CC BY-ND 4.0. - **Imply endorsement** by the LuMamba authors for any derivative or evaluation without our written permission. - **Use the LuMamba name** in a way that suggests your modified model is an official LuMamba release. ### ๐Ÿค How to contribute improvements (PR-gated releases) We welcome community improvements via a **pull-request (PR)** workflow. If you believe your improvements should become an **official LuMamba release**: 1. **Open a PR** in the [BioFoundation repository](https://github.com/pulp-bio/BioFoundation) describing the change (architecture/head/training recipe, datasets, preprocessing, compute). 2. Include **reproducibility artifacts**: configs, seeds, scripts, environment details, training/validation logs, and the **evaluation protocol** (e.g., TUAB/TUAR/TUSL) with exact splits. 3. Provide **comprehensive results** (AUROC/AUPR/BA, FLOPs, memory) vs. the baselines reported in the LuMamba paper. 4. After **maintainer review**, approved changes will be **retrained/validated** and, if accepted, **released by the maintainers** as a new **official LuMamba** checkpoint under **CC BY-ND 4.0**. > Rationale: CC BY-ND protects users from fragmented, lower-quality โ€œLuMamba variants,โ€ while still enabling internal fine-tuning and a path for the community to upstream improvements through review. --- ## ๐Ÿ”Ž Model Summary - **Goal:** Efficient and topology-agnostic EEG modeling with linear complexity in sequence length. - **Core idea:** **Channel-Unification Module** uses **learned queries** (Q) with **cross-attention** to map any set of channels to a fixed latent space. **bidirectional Mamba blocks** then operate on that latent sequence. - **Pre-training data:** TUEG, **>21,000 hours** of raw EEG; downstream subjects removed to avoid leakage. - **Downstream tasks:** **TUAB** (abnormal), **TUAR** (artifacts), **TUSL** (slowing), **SEED-V** (emotion; unseen 62-ch montage), **APAVA** (Alzheimer's disease; unseen 16-ch layout, **TDBrain** (Parkinson's disease; unseen 26-ch layout) --- ## ๐Ÿš€ Model Variants The model currently exists in a Tiny Variant, with the following parameters: | Variant | Parameters | FEMBA parameters |LUNA parameters | |-----------------|------------|-----------------------------|------------------------------------| | LuMamba_tiny | 4.1M |(`num_blocks` = 2, `exp` = 2)|(`num_queries` = 6, `embed_dim` = 64) Larger model sizes can be attained by increasing the number of bi-Mamba blocks `num_blocks` (e.g. 8 bi-Mamba blocks yields 15M parameters). --- ## ๐Ÿ“Š Results - **TUAB (abnormal vs normal):** 80.99 % Bal. Acc., 0.883 AUROC, 0.892 AUPR (LuMamba-Tiny, pre-trained with LeJEPA-reconstruction). - **TUSL (slowing event VS. seizure detection)**: 0.708 AUROC, 0.289 AUPR (LuMamba-Tiny, pre-trained with reconstruction-only). - **TUAR (artifact detection)**: 0.914 AUROC, 0.510 AUPR (LuMamba-Tiny, pre-trained with reconstruction-only). - **APAVA (Alzheimer's detection)**: 0.955 AUROC, 0.970 AUPR (LuMamba-Tiny, pre-trained with LeJEPA-reconstruction). - **TDBrain (Parkinson's detection)**: 0.961 AUROC, 0.960 AUPR (LuMamba-Tiny, pre-trained with LeJEPA-reconstruction). - **Mumtaz2016 (Depression detection)**: 0.725 Bal. Acc., 0.931 AUROC, 0.952 AUPR (LuMamba-Tiny, pre-trained with LeJEPA-reconstruction). - **SEED-V (5-class emotion detection)**: 0.350 Bal. Acc., 0.191 Cohen's Kappa (LuMamba-Tiny, pre-trained with reconstruction-only). - **MoBI (gait prediction)**: 0.116 R-squared, 0.148 RMSE (LuMamba-Tiny, pre-trained with reconstruction-only). - **MODMA (full 128-channel set)**: 59.47 % Bal. Acc., 0.448 AUROC, 0.420 AUPR (LuMamba-Tiny, pre-trained with reconstruction-only) - **MODMA (reduced 13-channel subset)**: 59.09 % Bal. Acc., 0.522 AUROC, 0.4153 AUPR (LuMamba-Tiny, pre-trained with LeJEPA-reconstruction). **Efficiency:** Up to **377ร— fewer FLOPs** relative to transformer-based baselines and supporting up to **500x longer** EEG windows, thanks to the efficient FEMBA bi-Mamba encoder. --- ## ๐Ÿง  Intended Use & Limitations **Intended use.** Research on EEG representation learning & classification (abnormality, artifacts, slowing, emotion), especially when **montages vary** or **channel counts are high**. **Limitations.** - **Not a medical device.** Do **not** use for clinical decisions without proper validation & regulatory clearance. - **Unseen topologies:** Zero-shot transfer to **very different/dense** layouts (e.g., SEED-V) can underperform SOTA despite positive scaling; consider augmenting pre-training montage diversity and spatial encodings. - **Distribution shifts:** Performance varies across cohorts, devices, and label protocols; validate locally and consider domain adaptation. --- ## ๐Ÿ—๏ธ Architecture & Training **LUNA Tokenizer & features.** EEG is patch-segmented; temporal features via 1D conv w/ GroupNorm+GELU; **frequency features** (FFT mag/phase โ†’ MLP) are added; 3D electrode coordinates encoded via **NeRF-style sinusoids โ†’ MLP** (positional enc). **LUNA Channel-Unification Module.** **Q learned queries** cross-attend to **channel-wise patch features** to produce a **fixed Qร—E latent** per patch; FFN + Transformer layers refine the query tokens. Complexity is **O(QยทC)** (linear in channels). **FEMBA Bi-Mamba Temporal encoder.** **Mamba blocks** process the embeddings in separate forward and backward streams. **Pre-training objectives.** **Masked-patch reconstruction** is used to reconstruct masked tokens. In parallel, the **LeJEPA loss** encourages an isotropic Gaussian embedding distribution to minimize downstream prediction risk. --- ## ๐Ÿ”ง How to Use LuMamba weights are organized by pre-training configuration: - **`Reconstruction-only`** โ†’ variants pre-trained with masked reconstruction exclusively - **`LeJEPA-reconstruction`** โ†’ variants pre-trained with a balanced mixture of masked reconstruction and LeJEPA losses. Variants exist for two different LeJEPA hyperparameters: 128 and 300 projection slices. - **`LeJEPA-only`** โ†’ variant pre-trained with LeJEPA exclusively. All variants are pre-trained on TUEG. LuMamba experiments are categorized by two Hydra configurations, in `BioFoundation/config/experiments`: - **`LuMamba_finetune.yaml`** โ†’ configuration for fine-tuning experiments. - **`LuMamba_pretrain.yaml`** โ†’ configuration for pre-training experiments. --- ## ๐Ÿ”ง Fine-tuning โ€” General Checklist 0. **Install & read data prep**: clone the [BioFoundation repo](https://github.com/pulp-bio/BioFoundation), set up the environment as described there, then open `make_datasets/README.md` for dataset-specific notes (naming, expected folder layout, and common pitfalls). 1. **Point to weights**: set `pretrained_safetensors_path: /path/to/LuMamba_*.safetensors` in the experiment YAML. 2. **Preprocess data**: acquire fine-tuning dataset and follow preprocessing protocol (see guide in `/make_datasets/README.md`) to generate `train/test/val.h5` files. 3. **Update data module of `LuMamba_finetune.yaml` config**: - **TUH datasets (TUAB/TUSL/TUAR)** โ†’ change `_target_` in `/data_module:` to `datasets.tuh_dataset.TUH_Dataset`. - **Other** โ†’ change `/data_module:_target_` to corresponding dataset.py file in `BioFoundation/datasets` (e.g., for TDBrain dataset use `_target_:datasets.tdbrain_dataset.TDBrain_Dataset`) - **HDF5 file location** โ†’ change `/data_module:hdf5_file` for `train`, `test`, and `val` with the path to the corresponding HDF5 data split file. 4. **Task settings**: - **Task type**: override with `/task:finetune_task_LUNA` for classification and `/task:finetune_regression_task_LuMamba` for regression tasks - **Classification type**: set `classification_type` (`bc`, `mcc`) and `model.num_classes` to match your downstream task. In a regression scenario,`mcc` is used and `model.num_classes` describes the number of features in the output. - **Classifier choice**: set `/model:classifier_option` (`mamba` for FEMBA classifier, `linear` for single-layer linear classifier,`null` for default LUNA classifier) - Configuration file includes further `#CHANGEME` tags and instructions for a working example. 5. **Env vars**: export `DATA_PATH` (dataset root) and `CHECKPOINT_DIR` (artifacts). 6. **Trainer/optimizer**: adjust `gpus/devices`, `batch_size`, `max_epochs`, LR/scheduler if needed. 7. **I/O**: set `io.base_output_path` and confirm `io.checkpoint_dirpath` exists. To launch fine-tuning (Hydra): ```bash python -u run_train.py +experiment=LuMamba_finetune ``` --- ## โš–๏ธ Responsible AI, Risks & Biases - **Clinical safety:** research-only; human oversight required. - **Bias & drift:** montage/device/population differences can induce shifts; validate and monitor. - **Artifacts & rare events:** robustness varies; use QC and task-appropriate preprocessing. --- ## ๐Ÿ”— Sources - **Code:** https://github.com/pulp-bio/BioFoundation - **Paper:** LuMamba: Latent Unified Mamba for Electrode Topology-Invariant and Efficient EEG Modeling (arxiv:2603.19100) --- ## ๐Ÿ“œ Citation If you use LuMamba, please cite: ```bibtex @misc{broustail2026lumambalatentunifiedmamba, title={LuMamba: Latent Unified Mamba for Electrode Topology-Invariant and Efficient EEG Modeling}, author={Danaรฉ Broustail and Anna Tegon and Thorir Mar Ingolfsson and Yawei Li and Luca Benini}, year={2026}, eprint={2603.19100}, archivePrefix={arXiv}, primaryClass={cs.AI}, url={https://arxiv.org/abs/2603.19100}, } ``` --- ## ๐Ÿ› ๏ธ Maintenance & Contact - **Issues & support:** please open a GitHub issue in the BioFoundation repository. --- --- ## ๐Ÿ”— Related Models - **[LUNA](https://huggingface.co/PulpBio/LUNA)** โ€” Transformer-based topology-agnostic EEG foundation model (NeurIPS 2025). Source of the channel-unification cross-attention module that LuMamba reuses. - **[FEMBA](https://huggingface.co/PulpBio/FEMBA)** โ€” Bidirectional Mamba foundation model for EEG. Source of the linear-complexity temporal backbone that LuMamba reuses. - **[TinyMyo](https://huggingface.co/PulpBio/TinyMyo)** โ€” Tiny foundation model for flexible EMG signal processing at the edge. ## ๐Ÿ—’๏ธ Changelog - **v1.0:** Initial release of LuMamba model card with task-specific checkpoints and instructions.