I am an assistant professor of Electrical and Computer Engineering and Computer Sciences (by courtesy) at the University of Wisconsin-Madison, a faculty fellow at the Grainger Institute, and a faculty affiliate with the Optimization group at the Wisconsin Institute for Discovery.
My research lies in the intersection of machine learning, coding theory, and distributed systems. I am particularly interested in the theory and practice of large-scale machine learning and the challenges that arise once we aim to build solutions that come with robustness and scalability guarantees.
Before coming to Madison, I spent two wonderful years as a postdoc at UC Berkeley, where I was a member of the AMPLab and BLISS, and had the pleasure to collaborate with
Ben Recht and Kannan Ramchandran.
I received my Ph.D. in 2014 from UT Austin, where I was fortunate to be advised by
Alex Dimakis. Before UT, I spent 3.5 years as a grad student at USC.
Before all that, I received my M.Sc. (2009) and ECE Diploma (2007) from the Technical University of Crete (TUC), located in the beautiful city of Chania.
In 2018, I co-founded the conference on Machine Learning & Systems (MLSys), a new conference that targets research at the intersection of systems and machine learning. In 2018 and 2020, I was the program co-chair for MLSys. In 2019, I also co-chaired the 3rd Midwest Machine Learning Symposium (MMLS).
My research has been supported by
an NSF CAREER Award,
two Sony Faculty Innovation Awards,
an AFRL Center of Excellence grant (MADLab),
an
ARPA-E Award, two
American Family Data Science Research Grants,
and
AWS Cloud Credits for Research.
gScholar
bio,
resume
twitter
Publications
2020
-
Optimal Lottery Tickets via SubsetSum: Logarithmic Over-Parameterization is SufficientNeurIPS 2020 (spotlight). [arxiv]
-
Attack Of The Tails: Yes, You Really Can Backdoor Federated LearningNeurIPS 2020. [arxiv]
-
Bad Global Minima Exist and SGD Can Reach ThemNeurIPS 2020 . [arxiv]
-
Closing the Convergence Gap of SGD without ReplacementICML 2020 . [arxiv]
-
Federated Learning with Matched AveragingICLR 2020 (oral). [openreview]
2019
-
DETOX: A Redundancy-based Framework for Faster and More Robust Gradient AggregationNeurIPS 2019. [arxiv]
-
Does Data Augmentation Lead to Positive Margin?ICML 2019. [arxiv]
-
Bad Global Minima Exist and SGD Can Reach ThemICML 2019, Deep Phenomena Workshop (oral). [arxiv]
-
A Geometric Perspective on the Transferability of Adversarial DirectionsAISTATS 2019. [arxiv]
2018
-
ATOMO: Communication-efficient Learning via Atomic SparsificationNeurIPS 2018. [arxiv]
-
The Effect of Network Width on the Performance of Large-batch TrainingNeurIPS 2018. [arxiv]
-
DRACO: Byzantine-resilient Distributed Training via Redundant GradientsICML 2018. [arxiv]
-
Stability and Generalization of Learning Algorithms that Converge to Global OptimaICML 2018. [arxiv]
-
Gradient Coding Using the Stochastic Block ModelISIT 2018. [arxiv]
-
Gradient Diversity: a Key Ingredient for Scalable Distributed LearningAISTATS 2018. [arxiv]
2017
-
Gradient Diversity: a Key Ingredient for Scalable Distributed LearningNIPS 2017, OPT Workshop (oral). [arxiv]
-
UberShuffle: Communication-efficient Data Shuffling for SGD via Coding TheoryNIPS 2017, ML Systems Workshop [pdf]
-
Perturbed Iterate Analysis for Asynchronous Stochastic OptimizationSIAM Journal on Optimization. [arxiv]
-
Speeding-up Distributed Machine Learning Using CodesIEEE Transactions on Information Theory. [arxiv]
-
Coded Computation for Multicore SetupsISIT 2017. [pdf]
2016
-
CYCLADES: Conflict-free Asynchronous Machine LearningNIPS 2016. [arxiv]
-
Speeding-up Distributed Machine Learning Using CodesISIT 2016. [arxiv]
-
On the Worst-Case Approximability of Sparse PCACOLT 2016. [arxiv]
-
Bipartite Correlation Clustering - Maximizing AgreementsAISTATS 2016. [arxiv]
-
Locality and Availability in Distributed StorageIEEE Transactions on Information Theory. [preprint]
-
Optimal Locally Repairable Codes and Connections to Matroid TheoryIEEE Trans. on Information Theory. [preprint]
2015
-
Perturbed Iterate Analysis for Asynchronous Stochastic OptimizationNIPS 2015, OPT Workshop (oral). [long version]
-
Speeding-up Distributed Machine Learning Using CodesNIPS 2015, ML-Systems Workshop. [arxiv]
-
Parallel Correlation Clustering on Big GraphsNIPS 2015. [long version]
-
Sparse PCA via Bipartite MatchingsNIPS 2015. [long version]
-
Orthogonal NMF through Subspace ExplorationNIPS 2015. [long version]
2014
-
Provable Deterministic Leverage Score SamplingKDD 2014. [long version]
-
Finding Dense Subgraphs via Low-Rank Bilinear Optimization
-
Nonnegative Sparse PCA with Provable Guarantees
-
Locality and Availability in Distributed StorageISIT 2014. [long version]
-
Locally Repairable CodesIEEE Transactions on Information Theory, September 2014.,
-
A Repair Framework for Scalar MDS CodesIEEE Journal on Selected Areas in Communications, May 2014.,
-
The Sparse Principal Component of a Constant-rank MatrixIEEE Transactions on Information Theory, March 2014. Megas., implementation by
2013
-
Sparse PCA through Low-rank Approximations
-
Optimal Locally Repairable Codes and Connections to Matroid TheoryISIT 2013. [IEEEXplore], [arXiv]
-
XORing Elephants: Novel Erasure Codes for Big Data
-
Repair Optimal Erasure Codes through Hadamard DesignsIEEE Transactions on Information Theory, May 2013. [IEEEXplore], [arXiv]
-
Maximum-Likelihood Noncoherent PAM DetectionIEEE Transactions on Communications, March 2013. [IEEEXplore], [draft]
2012
-
A Repair Framework for Scalar MDS CodesAllerton 2012. [IEEEXplore], [arXiv]
-
Locally Repairable Codes
-
Feedback in the K-user Interference channel
-
Simple Regenerating Codes: Network Coding for Cloud Storage
-
Maximum-likelihood Blind PAM DetectionICC 2012. [IEEEXplore], [long]
-
Interference Alignment as a Rank Constrained Rank Minimization
2011
-
Repair Optimal Erasure Codes through Hadamard Designs
-
Distributed Storage Codes through Hadamard Designs
-
Sparse Principal Component of a Rank-deficient Matrix
-
Repairing Erasure CodesUSENIX FAST 2011, Work-In-Progress (WiP). [short abstract]
2010
-
Distributed Storage Codes Meet Multiple-Access Wiretap Channels
-
MCMC Methods for Integer Least-Squares ProblemsAllerton 2010. [IEEEXplore], [draft]
-
Interference Alignment as a Rank Constrained Rank Minimization
-
Maximum-likelihood Noncoherent OSTBC Detection with Polynomial ComplexityIEEE Transactions on Wireless Communications, June 2010. [IEEEXplore], [draft]
2007-2009
-
Optimal OSTBC Sequence Detection over Unknown Correlated Fading ChannelsAsilomar 2009. [IEEEXplore], [draft]
-
Efficient maximum-likelihood noncoherent orthogonal STBC detection
-
Polynomial-complexity maximum-likelihood block noncoherent MPSK detectionICASSP 2008. [IEEEXplore], [draft]
-
Near ML detection of nonlinearly distorted OFDM signals