In my website you will find some things I have come to understand – by spending a ridiculous amount of time on them – and many more that confound me. Math, music, programming, photography and high-performance computing are things I know a tiny bit about and strive to learn more.

Research Interests

Machine Learning, High Dimensional Statistics, Large-scale Distributed Computation.

My work is motivated by the goal of making modern machine learning and data analysis accessible to non-experts and organizations with limited resources. Achieving this goal requires a number of theoretical and systems advances that will keep producing interesting research questions for years to come: fast and resource-efficient algorithms, self-tuning methods, data-dependent guarantees and distributed, large-scale computation are all necessary for a simple to use, automated ML pipeline.

My favorite problems span the intersection of theory and systems. Our recent work on asynchronous optimization identifies a previously unknown connection between optimization and system dynamics, with important implications for the setup and tuning of large-scale learning systems. More recently, I have been working on producing a self-tuning optimization algorithm for deep learning. I also work on inference, trying to understand the importance of and optimize the scan order used in Gibbs sampling.

In the past, I have worked on resource-limited problems like memory-limited streaming PCA as well as streaming PCA with overwhelming erasures in the entries of each sample. I have also worked on approximate but fast solutions for large-scale graph problems, like PageRank and finding dense subgraphs. Throughout our work I strive to bring together our very real implementation and the theory that will guarantee we get a good result.

Curriculumn Vitae

Detailed cv in pdf format.

Education

PhD in ECE, The University of Texas at Austin Austin, TX 2015.

MSc. in ECE, Technical University of Crete Greece, 2010.

Diploma, ECE, Technical University of Crete, Greece, 2008.

Advised by Prof. Nikos Sidiropoulos

Ask me about

How asynchrony brings about momentum

Does scan order affect Gibbs samplers?

How to optimize Gibbs for features

Fast PageRank on Graph Engines

Streaming PCA

Finding Dense Subgraphs of Large Graphs

Drop-d tuning

Shell vs Environment variables

Grilling the perfect lamb chop

Non-blocking I/O

Mic vs line levels

Slide Guitar

Side stitches

Tight hamstrings

The Python GIL

Western vs carnatic music

non-reversible markov chains

Wait-free parallelization