Research
Research I'm Chasing
These are the questions that keep me up at night. They sit in different fields, but they rhyme: each one is really about building machine learning you can trust when the data is incomplete and the answer matters.
Can a model tell a true story from a convincing one?
Misinformation is multimodal and it moves. I build models that read text and images together, model how a narrative drifts over time, and stay robust when they meet a dataset — or an event — they were never trained on.
What do brains and bodies reveal that words don't?
EEG and physiological signals carry honest traces of emotion, attention, and fatigue. My goal is recognition that generalizes across people, not just across recordings of the same person — the hard, deployment-relevant setting.
How do we model a disease we can only partially observe?
I combine mechanistic differential-equation models with Bayesian inference and machine learning to study tumor dynamics and nanoparticle-based therapy — quantifying uncertainty instead of hiding it.
Why did the model decide that — and should we believe it?
Across imaging and clinical data, I work on calibration, saliency, and uncertainty so a prediction comes with an explanation and an honest confidence, especially when a person's care is on the line.
For formal outputs, see my publications. For things I'm building and breaking, see projects & experiments.