AttnED

Using Encoder-Decoder with Attention Mechanism (AttnED) as the prediction model, and have three post-hoc XAI methods to choose from to explain the model - DiCE, LIME, and SHAP.

Using AttnED with SHAP is accepted at the AI4H Workshop at the 16TH INTERNATIONAL CONFERENCE ON SIGNAL IMAGE TECHNOLOGY & INTERNET BASED SYSTEMS as first author. Titled "Predicting and Explaining Hearing Aid Usage Using Encoder-Decoder with Attention Mechanism and SHAP".

Demo

demo.py shows how Attn_ED is used with the open sourced EvoSynth Data [Christensen et al. 2019].

Jeppe H. Christensen, Niels Pontoppidan, Rikke Rossing, Marco Anisetti, Doris-Eva Bamiou, George Spanoudakis, Louisa Murdin, Thanos Bibas, Dimitris Kikidiks, Nikos Dimakopoulos, & Apostolos Ecomomou. (2019). Fully synthetic longitudinal real-world data from hearing aid wearers for public health policy modeling (1.0: 08-04-2019: 4p) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.2668210

Citing AttnED

@inproceedings{su2022predicting,
  title={Predicting and Explaining Hearing Aid Usage Using Encoder-Decoder with Attention Mechanism and SHAP},
  author={Su, Qiqi and Iliadou, Eleftheria},
  booktitle={2022 16th International Conference on Signal-Image Technology \& Internet-Based Systems (SITIS)},
  pages={308--315},
  year={2022},
  organization={IEEE}
}