Awesome

Decision Metamamba

Overview

Decision MetaMamba Architecture: image info

A link of the paper can be found on arXiv.

Instructions

You need to install the metamamba library first using pip. Use the following command for installation:

cd path-to-DMM/metamamba
pip install -e .

After the installation is complete, you can train each dataset using the run.sh script.

Feature Map Activation for Proximal and Distal steps

Activation feature map after a Selective Scan SSM hidden state after ssm

Acknowledgements

Our Decision Metamamba code is based on decision-transformer decision-convformer.

and mamba

Citation

Please cite out paper as:

@misc{kim2024integratingmultimodalinputtoken,
      title={Integrating Multi-Modal Input Token Mixer Into Mamba-Based Decision Models: Decision MetaMamba}, 
      author={Wall Kim},
      year={2024},
      eprint={2408.10517},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2408.10517}, 
}

License

MIT