Grant: Discovery Grants Program - Individual
University of New Brunswick
$20,000.00
Geometric aspects of optimization
FY 2012-13
Summary
Many computational problems are either inherently geometric (e.g. CAD/CAM), or can usefully be analysed
in a geometric framework (e.g. the configuration space of robot). Of the latter type, applications in
e.g. pattern classification and non-parametric statistics are inherently ""high dimensional"", or concerned
with many more variables than our familar 3 or 4 coordinates. Computational geometry brings a rich
toolkit of algorithms and data-structures to these problems, but for the most part these are optimized
for the case of an enormous number of input objects in small dimensions. Techniques from optimization,
although typically slower for many objects in small dimensions, are usually more effective in the face of
many variables/dimensions. One theme from my research is thus the developement of better algorithms for
high dimensional geometric problems using techniques from optimization. Current projects in this theme
include training of maximum margin classifiers (support vector machines), representation transformation
for systems of linear inequalities, and computing the multivariate rank of a point in a data cloud.
A second theme in my work is the analysis of geometric structure as a way of understanding the behaviour
of algorithms, particularly algorithms in optimization. This analysis has two related aspects.
Convexity underlies almost all efficient optimization methods and has a rich geometric theory going back
to the 19th century. Combinatorial Geometry provides a way of understanding the structure of solution
sets, particularly important for discrete optimization algorithms. With several collaborators and using
a combination of theoretical tools and massive computation, I am currently analysing the structure of the
feasible regions of linear programs, particularly as it relates to the worst-case performance of the
simplex method.
Program
Discovery Grants Program - Individual
Recipient
Location
New Brunswick, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data
As published
Application id
228095-2010
Name nom
Bremner, David
Department département
Computer Science, Faculty of
Organization id
75
Institution établissement
University of New Brunswick
Province en
New Brunswick
Province fr
Nouveau-Brunswick
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2012
Competition year année de concours
2010
Award amount
20000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
Discovery grants programs
Group fr
Programmes de subventions à la découverte
Committee code
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
Area of application code
1207
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Information, computer and communication technologies
Area of application fr
Technologie de l'information, des ordinateurs et des réseaux de communications
Research subject code
2713
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Algorithms
Research subject fr
Algorithmes
Partie
3
Nb partie
5
Application title
Geometric aspects of optimization
Keyword
NSERC
Application summary
Many computational problems are either inherently geometric (e.g. CAD/CAM), or can usefully be analysed
in a geometric framework (e.g. the configuration space of robot). Of the latter type, applications in
e.g. pattern classification and non-parametric statistics are inherently ""high dimensional"", or concerned
with many more variables than our familar 3 or 4 coordinates. Computational geometry brings a rich
toolkit of algorithms and data-structures to these problems, but for the most part these are optimized
for the case of an enormous number of input objects in small dimensions. Techniques from optimization,
although typically slower for many objects in small dimensions, are usually more effective in the face of
many variables/dimensions. One theme from my research is thus the developement of better algorithms for
high dimensional geometric problems using techniques from optimization. Current projects in this theme
include training of maximum margin classifiers (support vector machines), representation transformation
for systems of linear inequalities, and computing the multivariate rank of a point in a data cloud.
A second theme in my work is the analysis of geometric structure as a way of understanding the behaviour
of algorithms, particularly algorithms in optimization. This analysis has two related aspects.
Convexity underlies almost all efficient optimization methods and has a rich geometric theory going back
to the 19th century. Combinatorial Geometry provides a way of understanding the structure of solution
sets, particularly important for discrete optimization algorithms. With several collaborators and using
a combination of theoretical tools and massive computation, I am currently analysing the structure of the
feasible regions of linear programs, particularly as it relates to the worst-case performance of the
simplex method.