Research

Representation is the language through which machines understand the world.

We investigate how meaningful representations emerge during optimization and how latent spaces capture structure, uncertainty, and abstraction.

Research topics include:

• Representation Learning
• Self-Supervised Learning
• Latent Variable Models
• Variational Autoencoders (VAEs)
• Probabilistic Generative Models
• Diffusion Models
• Foundation Models
• Mechanistic Interpretability
• Explainable AI

Human knowledge provides powerful inductive biases.

We study how biological knowledge, geometry, physical principles, and relational structures can be translated into mathematical formulations that guide learning.

Research topics include:

• Graph Neural Networks
• Probabilistic Generative Models
• Artificial Intelligence for Science (AI4Science)
• Implicit Neural Representations (INRs)
• Geometric Deep Learning
• Uncertainty Quantification

Modern AI systems increasingly perform reasoning rather than simple prediction.

We study how machines plan, retrieve knowledge, make decisions, and solve complex tasks through interaction with their environments.

Research topics include:

• Large Language Models
• Multimodal Learning
• Retrieval-Augmented Generation
• Agentic AI
• Reinforcement Learning
• Long-Context Reasoning

Scientific discovery provides one of the most challenging environments for machine learning.

Our laboratory develops AI methods for understanding biological systems and accelerating scientific research, while using these domains to study the principles of machine learning itself.

Current applications include:

• Computational Biology
• Gene Regulatory Networks
• Biomarker Discovery
• Multi-omics Integration
• Precision Medicine
• Biomedical Data Analysis

Rather than treating biology simply as an application, we view it as a rich source of hypotheses that inspire new machine learning methodologies.