Skip to content

Latest commit

 

History

History
35 lines (24 loc) · 2.03 KB

README.md

File metadata and controls

35 lines (24 loc) · 2.03 KB

bioFAME: Frequency-Aware Masked Autoencoders for Multimodal Pretraining on Biosignals

Publication link: Frequency-Aware Masked Autoencoders for Multimodal Pretraining on Biosignals.

Leveraging multimodal information from biosignals is vital for building a comprehensive representation of people's physical and mental states. However, multimodal biosignals often exhibit substantial distributional shifts between pretraining and inference datasets, stemming from changes in task specification or variations in modality compositions. To cope with this challenge, we propose a frequency-aware masked autoencoder bioFAME that learns to parameterize the representation of biosignals in the frequency space. bioFAME consists of a frequency-aware transformer and a frequency-maintain pretraining strategy. The resulting architecture effectively utilizes multimodal information during pretraining, and can be seamlessly adapted to diverse tasks and modalities at test time, remaining robust to distributional shifts. We evaluated our approach on a diverse set of transfer experiments on unimodal time series, achieving an average of 5.5% improvement in classification accuracy over the previous state-of-the-art. Furthermore, we demonstrated that our architecture is robust in modality mismatch scenarios, including unpredicted modality dropout or substitution, proving its practical utility in real-world applications.

Getting Started

Data preparation

  • Download the multimodal sleep staging dataset through the files within the data folder.
  • Download the finetuning datasets from the TFC codebase.

Perform pretraining

python -m main --data_dir YOUR-DATA-DIR --ft_data_dir YOUR-FT-DATA-DIR --reconstruction

Perform finetuning

python -m main --data_dir YOUR-DATA-DIR --ft_data_dir YOUR-FT-DATA-DIR --reconstruction --fine_tune --ft_new_data