Research Highlights

Our group builds machine learning methods for tabular and scientific data, with emphasis on efficient AI, interpretable feature selection, representation learning, multimodal learning, signal and audio analysis, and computational biology.

Structured sparsity method figure from the Train Less paper

Parameter-Efficient Training

We study low-rank optimization, structured sparsity, adaptive rank selection, and compression methods that make large models cheaper to train and deploy.

GroupFS method figure

Interpretability

Our interpretability work focuses on feature selection, sparse local explanations, and concept-level structure that remains useful in real scientific datasets.