2026-09-18T18:19:19Zhttps://ecommons.cornell.edu/server/oai/requestoai:ecommons.cornell.edu:1813/702252026-05-15T19:42:44Zcom_1813_35col_1813_47
Gheissari, Pargol
Huang, Yong
Estrin, Deborah
Azenkot, Shiri
2020-08-10T19:48:51Z
2020-08-10T19:48:51Z
2020-05
ProQuest Submission ID: 10838
ProQuest Publication ID: 27955778
ProQuest Submission ID: 10833
ProQuest Publication ID: 27955533
https://hdl.handle.net/1813/70225
https://doi.org/10.7298/v91k-j875
13254431
30 pages
In recent years, the amount of digital information stored in electronic health records (EHRs) has increased dramatically. At the same time, the advances in the field of machine learning, specifically deep learning has accommodated the opportunity for knowledge discovery and data mining algorithms to gain insight from this digital health data. Predictive modeling of clinical risks from EHRs, such as in-hospital mortality rate, in-hospital length of stay and chronic disease onset, can be helpful to the improvement of the quality of healthcare delivery. However, there are many challenges, such as sparsity, irregularity and temporality, associated with this clinical data. Therefore, this provides an opportunity for meta-learning methodologies to solve such problems and to have a large impact on medicine and quality of healthcare delivery. In this paper, we provide the background of this problem, review the commonly used strategies for solving such problems and discuss the state-of-the-art of meta-learning models. To address the clinical challenges associated with EHR data, we propose a meta-learning model, which uses latent-ODE as the base-learner and LSTM as the meta-learner, to solve disease phenotyping tasks. We then demonstrate that our proposed method outperforms the state-of-the-art models addressing classification tasks on healthcare data.
en
Meta-learning in Medicine
dissertation or thesis
https://catalog.library.cornell.edu/catalog/13254431
application/pdf
Information Science
Cornell University
Master of Science
M.S., Information Science
https://hdl.handle.net/1813/59810
Publication
Cornell Theses and Dissertations
Gheissari, Pargol
Huang, Yong
5893a6ea-7af3-41d7-abc6-04bcd26ab5df
open.access
open.access
<Publication xmlns="https://www.openaire.eu/cerif-profile/1.1/" id="03ccd3d5-7530-4cc4-b604-d6810d7e79db">
<Type xmlns="https://www.openaire.eu/cerif-profile/vocab/COAR_Publication_Types">http://purl.org/coar/resource_type/c_1843</Type>
<Language>en</Language>
<Title>Meta-learning in Medicine</Title>
<PublishedIn>
<Publication>
</Publication>
</PublishedIn>
<PublicationDate>2020-05</PublicationDate>
<DOI>https://doi.org/10.7298/v91k-j875</DOI>
<Authors>
<Author>
<DisplayName>Gheissari, Pargol</DisplayName>
<Affiliation>
<OrgUnit>
</OrgUnit>
</Affiliation>
</Author>
<Author>
<DisplayName>Huang, Yong</DisplayName>
<Affiliation>
<OrgUnit>
</OrgUnit>
</Affiliation>
</Author>
</Authors>
<Editors>
</Editors>
<Publishers>
<Publisher>
<OrgUnit />
</Publisher>
</Publishers>
<Abstract>In recent years, the amount of digital information stored in electronic health records (EHRs) has increased dramatically. At the same time, the advances in the field of machine learning, specifically deep learning has accommodated the opportunity for knowledge discovery and data mining algorithms to gain insight from this digital health data. Predictive modeling of clinical risks from EHRs, such as in-hospital mortality rate, in-hospital length of stay and chronic disease onset, can be helpful to the improvement of the quality of healthcare delivery. However, there are many challenges, such as sparsity, irregularity and temporality, associated with this clinical data. Therefore, this provides an opportunity for meta-learning methodologies to solve such problems and to have a large impact on medicine and quality of healthcare delivery. In this paper, we provide the background of this problem, review the commonly used strategies for solving such problems and discuss the state-of-the-art of meta-learning models. To address the clinical challenges associated with EHR data, we propose a meta-learning model, which uses latent-ODE as the base-learner and LSTM as the meta-learner, to solve disease phenotyping tasks. We then demonstrate that our proposed method outperforms the state-of-the-art models addressing classification tasks on healthcare data.</Abstract>
<Access xmlns="http://purl.org/coar/access_right"
>
</Access>
</Publication>