L3S Best Publication of the Quarter (1/2024)
Category: Time series modelling
IVP-VAE: Modeling EHR Time Series with Initial Value Problem Solvers
Authors: Jingge Xiao, Leonie Basso, Wolfgang Nejdl, Niloy Ganguly and Sandipan Sikdar
Presented at the 38th Annual AAAI Conference on Artificial Intelligence (A* Conference)
https://ojs.aaai.org/index.php/AAAI/article/view/29534
The paper in a nutshell:
Our paper introduces a novel approach to analyzing electronic health records (EHR). We propose a model that processes time series data through continuous processes parameterized by invertible neural networks, which simplifies model architecture and allows for parallel computation. This method enhances data efficiency and model performance.
Which problem do you solve with your research?
We address the challenge of efficiently processing irregularly sampled time series data commonly found in EHRs.
What is the potential impact of your findings?
The potential impact is beneficial for the healthcare industry. By improving the efficiency and accuracy of EHR data analysis, our model can aid in faster diagnosis, better patient monitoring, and personalized treatment plans.
What is new about your research?
The novelty lies in the use of neural initial value problem solvers within the variational autoencoder architecture for the modelling of EHR times series. This not only reduces the model’s complexity but also accelerates convergence, leading to state-of-the-art results in data efficiency and performance.
Xiao, J., Basso, L., Nejdl, W., Ganguly, N. and Sikdar, S. 2024. IVP-VAE: Modeling EHR Time Series with Initial Value Problem Solvers. Proceedings of the AAAI Conference on Artificial Intelligence. 38, 14 (Mar. 2024), 16023-16031. DOI: https://doi.org/10.1609/aaai.v38i14.29534.
