Welcome to the Latent Structure Lab! We build statistical tools for exploring latent variation in microbiological and earth systems data.

Modern sequencing, imaging, and spectroscopy technologies give a window into complex biological environments. However, exploring these data without getting lost remains a challenge — statistical methods can help chart out the territory. We are interested in building models and interfaces that help researchers summarize general patterns, spot interesting relationships, and articulate uncertainty. We have worked with probabilistic and latent variable models to better incorporate contextual knowledge and integrate complementary data sources.

• Latent variable modeling for the microbiome
• Multitable methods for microbiome data integration
• Interactive visualization of hierarchically structured data
• Bioconductor workflow for microbiome data analysis: from raw reads to community analyses

Climate change has affected many earth systems, and data about these changes can guide societal adaptation. Remote sensing instruments can provide detailed views of ecosystems at a global scale, but manual annotation and summarization is an overwhelming task. Features of interest can be few and far between, and designing models that work well across different environments remains a challenge. We adapt ideas from machine learning and data visualization to make remote sensing data useful for practical downstream tasks.

• Tackling climate change with machine learning
• Machine Learning for Glacier Monitoring in the Hindu Kush Himalaya
• Conditional Networks
• Application of Semantic Segmentation with Few Labels in the Detection of Water Bodies from Perusat-1 Satellite’s Images