
Michał Dereziński
Email: derezin at umich edu I am an Assistant Professor of Computer Science and Engineering at the University of Michigan. Previously, I was a postdoctoral fellow in the Department of Statistics at the University of California, Berkeley, and a research fellow at the Simons Institute for the Theory of Computing. I obtained my Ph.D. in Computer Science at the University of California, Santa Cruz, advised by professor Manfred Warmuth. My research is focused on the theoretical foundations of randomized algorithms for machine learning, optimization, and data science. Prior to UCSC, I completed Master's degrees in mathematics and computer science at the University of Warsaw. Research interests: Foundations of machine learning (ML), stochastic optimization, randomized numerical linear algebra (RandNLA), highdimensional statistics, and random matrix theory (RMT). To learn more about some of my interests, check out this tutorial on RandNLA for ML. 
Updates 

Teaching 
Current PhD students: Jiaming Yang, Sachin Garg

Publications 
Recent and Upcoming Developments in Randomized Numerical Linear Algebra for Machine Learning
Stochastic Newton Proximal Extragradient Method
Faster Linear Systems and Matrix Norm Approximation via Multilevel Sketched Preconditioning
Finegrained Analysis and Faster Algorithms for Iteratively Solving Linear Systems
Distributed Least Squares in Small Space via Sketching and Bias Reduction
Secondorder Information Promotes MiniBatch Robustness in VarianceReduced Gradients
HERTA: A HighEfficiency and Rigorous Training Algorithm for Unfolded Graph Neural Networks
Solving Dense Linear Systems Faster than via Preconditioning
Optimal Embedding Dimension for Sparse Subspace Embeddings
Surrogatebased Autotuning for Randomized Sketching Algorithms in Regression Problems
Algorithmic Gaussianization through Sketching: Converting Data into Subgaussian Random Designs
Randomized Numerical Linear Algebra  A Perspective on the Field with an Eye to Software
Sharp Analysis of SketchandProject Methods via a Connection to Randomized Singular Value Decomposition
Stochastic VarianceReduced Newton: Accelerating FiniteSum Minimization with Large Batches
Hessian Averaging in Stochastic Newton Methods Achieves Superlinear Convergence
Unbiased estimators for random design regression
Domain Sparsification of Discrete Distributions using Entropic Independence
NewtonLESS: Sparsification without Tradeoffs for the Sketched Newton Update
Query Complexity of Least Absolute Deviation Regression
via Robust Uniform Convergence
Sparse sketches with small inversion bias
LocalNewton: Reducing Communication Bottleneck for Distributed Learning
Determinantal Point Processes
in Randomized Numerical Linear Algebra
Debiasing Distributed Second Order Optimization with Surrogate Sketching
and Scaled Regularization
Sampling from a kDPP without looking at all items
Precise expressions for random projections: Lowrank approximation and
randomized Newton
Improved guarantees and a multipledescent curve for
Column Subset Selection and the Nyström method
Exact expressions for double descent and implicit regularization via surrogate random design
Isotropy and LogConcave Polynomials: Accelerated Sampling and HighPrecision Counting of Matroid Bases
Convergence Analysis of Block Coordinate Algorithms with Determinantal Sampling
Bayesian experimental design using regularized determinantal point processes
Exact sampling of determinantal point processes with sublinear time preprocessing
Distributed estimation of the inverse Hessian by determinantal averaging
Minimax experimental design: Bridging the gap between statistical and worstcase approaches to least squares regression
Fast determinantal point processes via distortionfree intermediate sampling
Correcting the bias in least squares regression with volumerescaled sampling
Leveraged volume sampling for linear regression
Reverse iterative volume sampling for linear regression
Subsampling for Ridge Regression via Regularized Volume Sampling
BatchExpansion Training: An Efficient Optimization Framework
Discovering Surprising Documents with ContextAware Word Representations
Unbiased estimates for linear regression via volume sampling
The limits of squared Euclidean distance regularization 