Gueorgui Mersov
Toronto, ON, M3H 5T9
416-***-**** (cell)
acb3ew@r.postjobfree.com
Objective: Development of statistical data treatment projects
(methods, algorithms, and programs).
Ph.D.: Artificial Intelligence in Space Navigation, Space
Research Institute, Moscow, 1980.
Data mining skills
. Ability to find new approaches to known problems and
realize it as a software tool.
. Deep understanding of Artificial Intelligence
methods, theory of probability and mathematical
statistics.
. Ability to understand specific features of the real
statistical data and to realize the needed data
mining procedure using traditional or ad-hoc created
algorithms.
Computer skills
C, C#, C++, SAS, Matlab.
Data mining experience:
From 2009 Chief Scientist, InferSystems Corp., Toronto
Development of predictive tools for a Real-Time Bidding.
A new approach to supervise learning is developed based on
Bernoulli's and Shannon theorems. The predictive model, which
is based on Bayes formula, can be built in automate mode (no
human intervention except specification the predicted event).
Computational time for model building does not exceed 1 hour for
historical data with millions of records and hundreds of
variables. The model can be used for RTB: computational time for
a prediction of the specified event for a given impression is
around 10 micro seconds.
2002-2009 Research scientist, Generation5 Mathematical Technology,
Toronto
Development of algorithms for time series prediction.
A new solution for ARIMA types models were proposed. It
permits to find an optimal autoregressive order using validation
approach. Developed algorithms programmed and debugged in C, C++
(Microsoft VS.net). High accuracy (from 1st to 3rd place among
24 methods published on WEB) and fast computation makes this
approach very competitive.
Development of clustering techniques based on estimated
probability density.
A new approach to clustering was proposed that based on
definition of clusters as a set of not connected subset of a
confidential set in multidimensional space. The corresponding
algorithms were programmed and debugged in Matlab environment.
Development of fast search of the nearest neighbor based on hash
coding.
Multidimensional set of records is divided on two
overlapping subsets of cells to solve boundary problem. The
records that are neighbors of a given record must be at any case
in the same cell (in the first or in the second subsets of
cells) with this record. To find these cells a special code
(hash value) was computed for all cells. This code is a multi
digitals integer with different base for every digit. Finally
the search of the nearest neighbors must be done only among
records with the same hash value. Computational time decreases
in tens times.
1980-2002 Development of control and predictive systems for
Aerospace projects in Russia and Israel.
Publications concerning different statistical problems in
predictive methods
1. Influence of Non Estimated Parameters on Estimation of Spacecraft
Coordinates (in Russian). Space Research Journal, V.10, No 1, 1982
2. Determination of Position and Attitude of Spacecraft on planet Mars (in
Russian). In the book Space research methods. Moscow, Nauka, 1982, pp.3-
15.
3. A Posteriori Algorithm for Filtering of Disturbing Parameters in
Autonomous Space Navigation (in Russian). In the book Applied Space
Navigation
Problems. Moscow, Nauka, 1983, pp 26-36.
4. Application of the Least Square Methods in Non Linear Space Navigation
(in Russian). In the book Software for Space Experiments. Moscow. Nauka,
1988, pp.63-68.
5 Algorithm for Localization of a Gamma Source Using 3 Satellites.
Astrophysics
and Space Science, V.75, 1981, pp. 31-34.
6. The Transformation of the Coordinates of a Gamma-Burst Source to the
Star Catalog. (Co-authors Bysnovaty-Kogan G.S., Estulin I.V. et al.)
Astrophysics and Space Science, V.75, 1981, pp. 213-224.
7. Precise Source Location of Anomalous 1979 March 5 Gamma-Ray Transient.
(Co-author Cline T.L., Dessai U.D., et al.) Astrophysics Journal
Letters, 256,1981,pp.L45-L48.
8. Einstein Observation of the 1978 November Gamma-Ray Source Field. (Co-
author Pizzichini G., Cline T., et al.) 17th ICRS, Paris, France, July,
xg2,2-2,1981.
9. Determination of Gamma-Burst Source Coordinates Using Time Delay
Measurements (in Russian). (Co-author Novak B.L.) In the book
Investigation of Gamma-Bursts on Space Stations. Moscow, Nauka, 1983,
pp.53-61.
10. Possibilities of Gamma Source Localization (in Russian). (Co-author
Estulin I.V.) Journal of Space Research, V.XX1, No 4, 1983, pp.88-90.
11. Robust Application of Least Square Method in Non Linear Space
Navigation (in Russian). In the book Navigation in Space Experiments.
Moscow, Nauka, 1984, pp.127-150.
12. Statistical Approach to the Gamma-Burst Sources Localization (in
Russian). In the book Problems and Methods of Space data treatment.
Moscow, Nauka, 1987, pp.122-126.
13. The Relativistic Effects in Localization of Gamma-Burst Sources. In the
book Relativity in Celestial Mechanics and Astrometry. D.Reidel
Publishing Company, Dordrecht, Holland, 1986, pp.215-222.
14. Estimation of Spacecrafts Attitude Using Angular Measurements (in
Russian). Journal of Space Research, V.XXI, No 6, pp.840-860.
15. Can We Expect a Freely Precessing Neutron Star in Her-X1? (Co-author
Bysnovaty-Kogan G.S, Shefer E.K.) Astronomy&Astrophysics, 221, 1989,
pp. L7-L9.
16. Investigation of Evolution of the Non Linear System Disturbed by Noise.
(Co-
author Moiseev S.S., Nezlin Y.M.) Journal of Chaos, 1993.
17. The prediction of stellar UV magnitudes.(Co-author Brosch N.) 1993 IPS
meeting, Tel Aviv University, April 1993.
18. The prediction of stellar UV colors. (Co-authors Shemi A., Brosch, N.,
Amoznino, E.) ADASS Conference, Victoria, BC., Canada, October 1993.
A.S.P. Conference Series, Volume 61. "Astronomical Data Analysis
Software and Systems III", ed. Crabtree, D.R., Hamish, R.J. and Barnes,
J.