Research philosophy

Learning what to share.
Preserving what matters.

My work asks how learning systems can use relationships, transfer information across domains, and incorporate biological knowledge without losing meaningful differences.

Direction 01

Graph representation learning

Learning from relationships, not just individual observations. I study graph-based and semi-supervised methods that use network structure when labels are scarce.

How can multiple relationships support learning when supervision is limited?

Direction 02

Explainable biomedical AI

Connecting neuroimaging, plasma proteomics, and biological knowledge to study disease heterogeneity and interpretable biomarkers.

How can biological structure make multimodal predictions more interpretable?

Direction 03

Learning across domains & time

Transferring information between heterogeneous populations, measurement settings, and time points while preserving useful differences.

What should be shared—and what should be preserved—across domains?

An organizing perspective

From method to biomedical question.

01 / REPRESENT

Use the structure

Represent relationships among observations and biological entities rather than treating every feature in isolation.

02 / TRANSFER

Respect the differences

Study which information can be shared across populations, modalities, and observation times.

03 / INTERPRET

Connect back to biology

Relate model representations to disease heterogeneity, biological pathways, and candidate biomarkers.

Learning what to share.Preserving what matters.