Towards Dynamic Feature Acquisition on Medical Time Series by Maximizing Conditional Mutual Information
Abstract
Knowing which features of a multivariate time series to measure and at what time is a key task in medicine, wearables, and robotics. Better acquisition policies can reduce costs while maintaining or even improving the performance of downstream predictors. Inspired by the maximization of conditional mutual information, we propose an approach to train acquirers end-to-end using only the downstream loss. We show that our method outperforms random acquisition policy, is close to the performance of a model with an unrestrained budget, but can't match a static acquisition strategy (likely due to the simplicity of its architecture). We highlight the assumptions and outline avenues for future work.
Cite
Text
Sergeev et al. "Towards Dynamic Feature Acquisition on Medical Time Series by Maximizing Conditional Mutual Information." ICML 2024 Workshops: SPIGM, 2024.Markdown
[Sergeev et al. "Towards Dynamic Feature Acquisition on Medical Time Series by Maximizing Conditional Mutual Information." ICML 2024 Workshops: SPIGM, 2024.](https://mlanthology.org/icmlw/2024/sergeev2024icmlw-dynamic/)BibTeX
@inproceedings{sergeev2024icmlw-dynamic,
title = {{Towards Dynamic Feature Acquisition on Medical Time Series by Maximizing Conditional Mutual Information}},
author = {Sergeev, Fedor and Malsot, Paola and Ratsch, Gunnar and Fortuin, Vincent},
booktitle = {ICML 2024 Workshops: SPIGM},
year = {2024},
url = {https://mlanthology.org/icmlw/2024/sergeev2024icmlw-dynamic/}
}