Mikhail Yurushkin

Rostov on Don, Rostov, Russian Federation

Yearly Stats: $0 Earned |

PhD, deep learning/data scientist

My homepage with relevant CV: yurushkin.ru

Machine learning skills
• Instruments: tensorflow/serving/keras, lasagne/theano, python/matlab, numpy/pandas/matplotlib, RapidMiner/Weka
• Methods: gradient boosting/random forest/svm/linear regression/ridge regression/etc
Commercial projects in which I successfully used machine learning:
• I successfully implemented algorithm of windows drivers ranking (based on weka random forest). It’s used by Carambis Driver Updater utility.
• High frequency trading for russian exchange. My own startup. This is a most di?cult my project. Still in progress...
• Prediction of city energy consumption. I used matlab/fuzzy neuro network to solve this problem.
• Products classi?cation by categories. I automatically classi?ed big number of products by categories for o?erbox.ru. Prototype is implemented in python. High performance algorithm is implemented in pure C++/boost.
• Optimization of weka random forest in C++.

High Performance computing, program optimization and compilers development
My workspace: gcc/icc/vtune/qtcreator/linux.
I have experience in the following areas:
• low level optimization: avx vectorization, pipelining, data prefetching, loop unrolling, data cache, virtual memory/tlb cache.
• parallel computing (openmp/pthreads, mpi/hadoop/apach spark). I’ve been working since 2008 year in the group of programmers and scientists. We develop multi-front-end optimizing compiler - Optimizing Parallel System (OPS).
My projects:
• FORTRAN 77/90 front-end implementation in OPS project. I used C++/Antrl/Rose compiler.
• Front-end for stack oriented language implementation (my Intel internship project).
• Middle-end: block data placement automatization support in C compiler. I implemented directives which help programmers to use block data placement in real programs.
• High performance matrix multiplication algorithm (DGEMM) with double blocking in shared memory. My implementation of this algorithm has better performance than Intel MKL, PLASMA, OpenBLAS.

Native development skills
I have solid (7+ years) experience of non-stop cross-platform native C++ development for windows/linux/freebsd.
• Main languages: C++, python, NSIS.
• Compilers: gcc/icc/clang. • Frameworks: boost/qt/qml/gstreamer/curl/zlib/libtorrent/htmlayout.
• other: redmine/trac, git/svn, doxygen.

$40 / Hour
$1,000 minimum budget