---
license: mit
pipeline_tag: text-classification
---

# BarcodeMamba+: Advancing State-Space Models for Fungal Biodiversity Research

BarcodeMamba+ is a foundation model for fungal barcode classification, built on a powerful and efficient state-space model architecture. It addresses critical challenges in fungal taxonomic classification, such as sparse labelling and long-tailed taxa distributions, by employing a pretrain and fine-tune paradigm. The model integrates various enhancements, including hierarchical label smoothing, a weighted loss function, and a multi-head output layer, to achieve significant performance gains over traditional supervised methods.

- Check out our [paper](https://huggingface.co/papers/2512.15931)
- Check out our [code](https://github.com/bioscan-ml/BarcodeMamba)
- Check out our [poster](https://neurips.cc/media/PosterPDFs/NeurIPS%202024/105938.png)

# Usage
The pretrained models can be used for both taxonomic classification on seen species (fine-tune & linear probe) and making genus-level predictions on unseen species (1-NN probe). The instructions for using our models can be found at our [GitHub repository](https://github.com/bioscan-ml/BarcodeMamba).

# Citation

If you find BarcodeMamba useful, please consider citing:

```
@inproceedings{
gao2024barcodemamba,
title={BarcodeMamba: State Space Models for Biodiversity Analysis},
author={Tiancheng Gao and Graham W.~Taylor},
booktitle={{NeurIPS} 2024 Workshop on Foundation Models for Science: Progress, Opportunities, and Challenges},
year={2024},
url={https://openreview.net/forum?id=6ohFEFTr10}
}
```