Ph.D. student, Computer Science, University of California, Santa Cruz.
M.S. in Computer Science, University of Warsaw.
Thesis: On Generating Concept Hierarchies with Fuzzy Data
PDF
M.S. in Mathematics, University of Warsaw.
Thesis: Isomorphic Properties of Function Space BV on Simply Connected Planar Sets
PDF
Leveraged volume sampling for linear regression
M. Dereziński, M. Warmuth, D. Hsu
2018
arXiv
Proposed a new variant of volume sampling, which produces the first unbiased least squares estimator with strong loss bounds for linear regression. Also, proposed a new faster joint sampling strategy called determinantal rejection sampling, and developed new techniques for proving tail bounds for the sums of dependent random matrices that arise from volume sampling.
Reverse iterative volume sampling for linear regression
M. Dereziński, M. Warmuth
2018
arXiv
Introduced the reverse iterative volume sampling framework, used to develop new expectation formulas (e.g. unbiasedness of volumesampled least squares estimator), a regularized extension of volume sampling with statistical guarantees, and stateoftheart sampling algorithms.
Subsampling for Ridge Regression via Regularized Volume Sampling
M. Dereziński, M. Warmuth
AISTATS 2018
arXiv
Poster
Talk
Proposed a new efficient procedure for solving a ridge regression task over a large unlabeled dataset,
when computing labels is expensive. This method, called regularized volume sampling, is fast,
easy to implement, and offers strong statistical guarantees.
BatchExpansion Training: An Efficient Optimization Framework
M. Dereziński, D. Mahajan, S. S. Keerthi, S.V.N. Vishwanathan, M. Weimer
AISTATS 2018
arXiv
Poster
Talk
Proposed a parameterfree batch optimization method running on a gradually
expanding dataset, which outperforms standard batch
and stochastic approaches in largescale settings.
Discovering Surprising Documents with ContextAware Word Representations
M. Dereziński, K. Rohanimanesh, A. Hydrie
IUI 2018
PDF
Talk
Developed a informationtheoretic model for discovering surprise in text, that outperforms deep learning in a practical task.
Unbiased estimates for linear regression via volume sampling
M. Dereziński, M. Warmuth
NIPS 2017
arXiv
Poster
Spotlight
Showed that volume sampling provides an unbiased estimate of the least squares solution
from a subset of labels, which is close to the optimum solution up to a multiplicative factor.
Also, proposed an efficient algorithm for volume sampling.
Anticipating Concept Drift in Online Learning
M. Dereziński, B. N. Bhaskar
LFED Workshop at NIPS 2015
PDF
Developed and analyzed online optimization algorithms
that dynamically adapt to concept drift.
The limits of squared Euclidean distance regularization
M. Dereziński, M. Warmuth
NIPS 2014
PDF
Spotlight
Proved a lower bound for the loss of learning algorithms that regularize with squared Euclidean distance, for a certain family
of learning problems. Proposed and experimentally tested a conjecture extending the result to deep learning.
Active SemiSupervised Concept Hierarchy Refinement
M. Dereziński
LAWS 2012 workshop
PDF
Developed and tested a new algorithm for refining concept
hierarchies generated from unsupervised data, by actively
querying a human expert.
