Research at SOL

Designing new methodology to learn the underlying structure of messy data — with missing values, imbalanced classes, nonnegativity restrictions, limited sample sizes, and beyond.

Our goal is to understand when this type of learning is possible (theoretical conditions), and how to learn these structures efficiently (practical algorithms).

We use a unique integration of statistics, optimization, algebraic geometry, and machine learning. SOL's current research topics include:

Nonnegative Matrix Factorization

Breaking down complicated data into a small set of building blocks that reveal how it works. The added twist is that we cannot use negative numbers.

Neonatal Encephalopathy

Predicting a newborn brain disorder using Machine Learning. The challenge is that data is very scarce.

Visualizing Spaces of Structures

Designing tools to visualize the world of high-dimensional structures — 4D and far beyond.

Characterizing Metastasis

Uncovering the drivers of metastasis using clinical data, RNA sequences, and Machine Learning.

Understanding AI

Artificial intelligence (AI) is making a big splash. Yet nobody really knows how it works! We want to change that.

Contrastive Learning

Recognizing what is similar and what is different.

Subspace Clustering when Data is Missing

Grouping data points that follow the same linear pattern. The challenge is that data is highly incomplete.

Conservation and Biodiversity

Determining the health of an ecosystem through nature's soundtrack.

Multimodal Translation

Is it possible to translate between RNA sequences and medical images?