Maxim Raginsky: Research Themes and Selected Publications

Click here for the complete list of my publications, with full citation data and links to electronic versions.

Compressed representations for statistical learning and inference

I am interested in developing theory and algorithms for efficient data representations for transmission and storage that reduce bandwidth and memory requirements as much as possible, yet preserve most relevant features needed for inference. Applications include field estimation using wireless sensor networks, smooth binary codes for fast similarity search and retrieval, and design of informative dictionaries for scene classification in computer vision.

Connections between information theory, estimation, and optimization

I am interested in new angles and perspectives at the interface between information theory, estimation, and optimization. Applications include improving the complexity of convex optimization using information-theoretic coding, efficient methods for sequential decision making and anomaly detection, and universal schemes for information processing that perform nearly optimally even with incompletely or inaccurately specified system models.

Nonparametric methods for discrete data

Many current applications involve decision-making and inference from discrete event data. The main challenge is when the number of possible event types is much larger than the number of observed events. I am interested in developing "nonparametric" inference methods for such settings. Applications include compressed sensing based on photon counts, sparse density estimation on the binary cube, and discrete denoising for region classification in computer vision.