Michal Derezinski
University of Michigan, MI·Computer Science·Early-career
Nearly every algorithm in modern data science eventually reduces to arithmetic on matrices, the grids of numbers used to store data. Doing that arithmetic exactly becomes impractical as datasets grow, so algorithm designers turn to randomness: they compute with small random summaries of the data rather than all of it.
- Fields
Computer Science
Mathematics & Statistics
- Topics
randomized numerical linear algebra
RandNLA
matrix sketching
- Grants
- 1 known grant, through Apr 2029
What they work on
- Reformulate optimal matrix sketching, the construction of a small randomized approximation of input data, using black-box sampling methods, so that sketch and sub-sample outputs come with reliability guarantees suitable for software.
- Build randomized iterative refinement algorithms that solve linear systems and quadratic problems to high precision, working in the regime between deterministic iterative solvers and stochastic optimization.
- Analyze the robustness of randomized numerical linear algebra algorithms and identify matrix approximation schemes that preserve data structure for feature selection, experimental design and interpretability.
- Supply the algorithmic and theoretical foundations needed to incorporate randomization into LAPACK, the standard computational linear algebra package used in scientific computing.
More about the research
These randomized methods are fast in principle, but their behavior is hard to control in practice, and that unreliability has kept them out of widely used software. Narrowing this theory-to-practice gap is the purpose of this work.
Why it matters
Matrix computations sit underneath analyses used in healthcare, climate research and urban planning, so more dependable randomized algorithms would make those analyses cheaper at large scale. Because the project aims at integration with LAPACK, improvements could reach the many scientists, engineers and machine learning practitioners who use that library without writing numerical code themselves.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Vivak Patel
University of Wisconsin-Madison, WI·Mathematics & Statistics
Map a modular theory of randomized linear algebra algorithms onto composable computational kernels so that new, application-tailored randomized solvers can be assembled rapidly rather than coded from scratch.
Anna Ma
University of California-Irvine, CA·Mathematics & Statistics
Analyze the sample complexity of low-rank tensor completion, meaning how many observed entries are needed to recover a multi-dimensional dataset with simple underlying structure.
Stephen Becker
University of Colorado Boulder, CO·Mathematics & Statistics
Develop gradient-free optimization algorithms for objective functions whose matrix-valued gradients have low intrinsic dimensionality, such as sparsity, low rank, or sparsity-plus-low-rank, to break worst-case efficiency barriers.
The school
University of Michigan: 124 labs on PI Pages, $2.1B a year on research, 963 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.