Christopher Musco Research Group

The group develops randomized linear-algebra sketching methods to compress matrices and accelerate large-scale machine learning while preserving provable error bounds. The group designs streaming and sketching algorithms that process high-rate data with sublinear memory, enabling one-pass summaries for statistical estimation. Researchers analyze optimization and numerical linear algebra mechanisms to improve solver convergence and stability for large problems encountered in data science. The team implements algorithmic prototypes and benchmarks to evaluate tradeoffs between runtime, memory, and theoretical guarantees.