CURRICULUM VITAE

			George Buyanovsky

Date of Birth:		September 28, 1964			
Address:		CIS, Republic of Kazakhstan
			Almaty, 480063,
			Microregion Zhetisu-2, 56-9
			Internet E-mail: george@acb.alma-ata.su
			fax:7_3272_623856  tel:7_3272_271317

Education:	1) Almaty Polytechnical College, Diploma	1982-1987
		2) Secondary School, Certificate		1972-1982

Working experience:

November 1993- 		Chief specialist of the Department of Perspective
-Present		Developments of the Joint-Stock Company
			KRAMDS-INFORMATION SYSTEMS: data compression,
			neural- network modelling for the problems of
			prognosis of behaviour of multifactor systems and
			problems of images recognition, author of original
			algorithms and programs
			
September 1991-		Joint-Venture "SovAvstralTechnica"
-November 1993		system programmer, data compression, prognosis of
			dynamic series, stereographics, local networks,

November   1989-	Almaty Subsidiary of Joint-Venture "Interquadro",
-September 1991		mathematician-programmer problems of discrete
			optimization, location task and task on covering,

September  1987-	Kalchugin Plant of Non-Ferrous Metals
-November  1989		engineer-programmer of the first category

	Data Compression _ ACB-compressor (1994-1995):
worked out a new algorithm of data compression ACB-compressor
(Associative Coder of Buyanovsky). The archivator of general purposes
ACB.EXE. is realized. The size of  ACB-archives is by 10-95% less relative
to sizes ZIP- archives (PKZIP V2.04g).  The algorithm works with data stream,
suitable for communication purposes and apparatus realization. Please, see
ACB.txt for detailed information on ACB-compressor with the program
ACB.EXE Ver_1.13b.

	Prognosis of Dynamic Series (1994):
the idea of the algorithm is based on the organization of associative memory
with the prognosis by the funnel of analogies. The funnel of analogies is
built on the data up to 7 exponent with rounding on each of the series
(256 128 64 32 16 8 4 2) of the discretization levels.  The program is
written under  Windows 3.1. For testing the generator of dynamic series is
included in the program. The effectiveness of the program was compared with
the package "Mezozavr" of Aivozyan' s group, at exchange rates relative to
Dutch crone. The prognosis was considered successful if fall/rise was
predicted by the following counting. "Mezozavr" showed 52/48, the tested
algorithm  55/45  per cent. I also worked out an algorithn of quick sorting
of bit vectors, with time complexity 3N+o(1) on the data "white noise".

The previous two algorithms are based on my articles
"The method of research of pseudostochaistic systems" published in the annals
of the Institute Of High Energy Physics of the Academy of Sciences of the
Republic of Kazakhstan in 1993 in English and the article "Associative coding"
in the 8-th issue of the magazine "Monitor " , 1994 .

	Neural-Network Modeling of multifactor systems (1994):
the program of prognosing the behaviour of multifactor systems with a
possibility of revealing mutual reasons of factors was worked out. Weight
input coefficients of neurons can be received in the size of teaching
selection not less n+2 calculating, where n- number of factors of the
modelled system. Time complexity of calculating weight coefficients
P(o)=4/3*n*n*n. After rate fixing weight input coefficients of neurons can be
interpreted as coefficients of reasons and can be used for the analysis of
the logical structure of the system. After calculating the model (weight
coefficients)  the model can be started for self-development from any
starting condition of the system (what happens if...). The program works in
MS_WINDOWS_3.1 (win32s), exchanging input and output data through Clipboard
or through text files. Supports DDE interface. Formation of initial data can
be made in  Excel,Lotus and etc. On computers with the processor i486-33mHz.
RAM 8 Mbyte it is possible to model systems with sizes up to 1000 factors in
time of calculating the model- 20..30 min  and the prognosis for the step -
- 3..6 sec.)

	Images recognition (1992-1994):
the program of recognition of the belonging of a graphic image to one of the
classes on the tree of classes was worked out. The algorithm of building a
tree of logical utterances in the space of indicators, received with the help
of Furie-images on fractal orders, is used. The peculiarities of the
algorithm: in global teaching an hierarchical neural network N1 is built for
all classes. In recognizing on N1 on the received subset of candidates a
neural network N2...N3... and etc. is built until receiving one-element set.
On HD only N1 is saved. The program works under WINDOWS 3.1+.
There are two regimes:
expert		 - interactive regime of replenishing teaching succession,
		education of the system and recognition of the classification
		of new images for sudordinate lists of classes;
user's		 - applied program places a graphic image 
		into "clipboard" WINDOWS reads from it the answer( route on
		the tree of classes), that allows to expand possibilities of
		the existing systems of data base management to search
		information by key - graphical image 
Possibilities:
 1) Maximal number of classes in one knot of the tree of classes 1200.
 2) Debth of the tree of classes is restricted by the capacity of Winchester.
 3) Graphic image - black and white.
 4) Number of pixels of the map of a graphical image - not more than 65536.
 5) Number of pixels forming an image - not more than 8192.

Time complexity of the algorithm of recognition is described in integer
operations by the following formulae:
) teaching	-  N2MKLog2(K)+o(1),
) recognition	-  NLog2(K)+CN'2MK.
Expenditure of memory (byte): RAM - 4NMK+2NK, HD - 8NMK+o(1).
where:	N - number of classes;
	N' -subset N candidates (N'<<N) ;
	M -number of indicators;
	K -average size of selection by class;
	 -number of iterations of bringing to one candidate (<<N').

	Location problem (1989-1991):
NP-complex problem. To solve this task I used Dinitz' idea on correcting
algorithms.  The principle of correcting algorithms is the following:  for
any NP-complex task there exist polynomial particular cases, the solution
received on the data corrected to one of them differs from the global
extremum by not more than the sum of all the corrections, that allows to
realize the branch-and -bound method. For solution of the location task data
on which Cherenin's "Preliminary Rejection Algorithm" worked up to the global
extremum were taken as a polynomial case.  The branch-and-bound method was
built on the basis of this regularity. Using Berestnev's invariants linear
time complexity for every knot of the tree was achieved. The program was
written in 1989 and allows to process real data with the dimensions 100100
within half an hour on  IBM_AT 286-12 gz.

	Stereographics (1991):
educational program under Windows 3.1 on the course of analitical geometry
for visualization of stereoscopic images of three-dimensional  functions in
spherical and Decart coordinates. On a model of a three-dimentional image in
the regime of real time students choose coordinates of observation,
perspective, stereobase, lights, after that full three-dimentional picture of
the function is formed. The surface of the fucntion can me nontransparent,
semitransparent, latticed. It is possible to select colors for left and right
images according to available color glasses. Time for making a picture
(counting up to 10000) model - momentarily, full picture  5 -10 sec on IBM_AT
386-40/387.

Computer skills:	Assembler 80386 (real_16/protection_32),
			 (Symantec C++_v6.1), SDK_MS_WINDOWS 3.0+ &(win32s)
