Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

University of Waterloo

$13,000.00

Flexible Angle-based Classification with High Dimensional Data

FY 2015-16

Summary
Classification is a commonly used statistical tool for information extraction. Many classifiers have <span style=""line-height: 1.5em;"">been proposed and extensively used for data sets with many observations and a few predictors. Recent advances <span style=""line-height: 1.5em;"">in technology, however, have enabled scientists to collect massive amounts of data with very high <span style=""line-height: 1.5em;"">dimension. Extracting knowledge from such large data sets poses unprecedented challenges to existing <span style=""line-height: 1.5em;"">learning techniques. In the last few decades, machine learning based classifiers have become increasingly <span style=""line-height: 1.5em;"">popular, partly due to their ability to handle massive data sets; but despite the current successes, many <span style=""line-height: 1.5em;"">open problems remain. For example, in simultaneous multicategory classification, a common approach <span style=""line-height: 1.5em;"">is to use a classification function vector, whose length is the same as the number of classes. To reduce <span style=""line-height: 1.5em;"">the parameter space and ensure some theoretical properties, a sum-to-zero constraint is often imposed <span style=""line-height: 1.5em;"">on the classification function vector. This formation can be redundant and inefficient. A new anglebased <span style=""line-height: 1.5em;"">classification framework was recently introduced to overcome this difficulty, and was shown to <span style=""line-height: 1.5em;"">be highly competitive for high dimensional classification problems. The existing angle-based classifiers <span style=""line-height: 1.5em;"">aim to solve standard classification problems. This research program will develop novel machine learning <span style=""line-height: 1.5em;"">methodologies to generalize and adapt the angle-based classification framework to tackle problems <span style=""line-height: 1.5em;"">where the primary learning goal is not simply class label prediction.  <span style=""line-height: 1.5em;""><span style=""line-height: 1.5em;"">The proposed work will focus on two aims. Aim 1 is to construct novel machine learning methods <span style=""line-height: 1.5em;"">when the learning goal is more than mere class label prediction, such as class conditional probability <span style=""line-height: 1.5em;"">estimation. In Aim 2, I plan to tackle problems where the observations are not paired. For example, <span style=""line-height: 1.5em;"">in semi-supervised learning, only a small proportion of observations have their labels available. The <span style=""line-height: 1.5em;"">proposed study will have significant value in probability, machine learning and non-parametric statistics. <span style=""line-height: 1.5em;"">For HQP training, Master's and PhD students will be intensively involved in the research process. <span style=""line-height: 1.5em;"">They will gain solid experience in developing efficient algorithms, conducting numerical analysis, and <span style=""line-height: 1.5em;"">collaborating with scholars in many other disciplines.
Program
Discovery Grants Program - Individual
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2015-04673
Name nom
Zhang, Chong
Department département
Statistics and Actuarial Science - Statistics and Actuarial Science
Organization id
33
Institution établissement
University of Waterloo
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2015
Competition year année de concours
2015
Award amount
13000
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
1508
Committee name en
Mathematics and Statistics
Committee name fr
Mathématiques et statistique
Area of application code
1205
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Mathematical sciences
Area of application fr
Sciences mathématiques
Research subject code
3004
Research subject group en
Statistics and probability
Research subject group fr
Statistique et probabilité
Research subject en
Applied statistics
Research subject fr
Statistique appliquée
Partie
1
Nb partie
5
Application title
Flexible Angle-based Classification with High Dimensional Data
Keyword
NSERC
Application summary
<p>Classification is a commonly used statistical tool for information extraction. Many classifiers have <span style=""line-height: 1.5em;"">been proposed and extensively used for data sets with many observations and a few predictors. Recent advances </span><span style=""line-height: 1.5em;"">in technology, however, have enabled scientists to collect massive amounts of data with very high </span><span style=""line-height: 1.5em;"">dimension. Extracting knowledge from such large data sets poses unprecedented challenges to existing </span><span style=""line-height: 1.5em;"">learning techniques. In the last few decades, machine learning based classifiers have become increasingly </span><span style=""line-height: 1.5em;"">popular, partly due to their ability to handle massive data sets; but despite the current successes, many </span><span style=""line-height: 1.5em;"">open problems remain. For example, in simultaneous multicategory classification, a common approach </span><span style=""line-height: 1.5em;"">is to use a classification function vector, whose length is the same as the number of classes. To reduce </span><span style=""line-height: 1.5em;"">the parameter space and ensure some theoretical properties, a sum-to-zero constraint is often imposed </span><span style=""line-height: 1.5em;"">on the classification function vector. This formation can be redundant and inefficient. A new anglebased </span><span style=""line-height: 1.5em;"">classification framework was recently introduced to overcome this difficulty, and was shown to </span><span style=""line-height: 1.5em;"">be highly competitive for high dimensional classification problems. The existing angle-based classifiers </span><span style=""line-height: 1.5em;"">aim to solve standard classification problems. This research program will develop novel machine learning </span><span style=""line-height: 1.5em;"">methodologies to generalize and adapt the angle-based classification framework to tackle problems </span><span style=""line-height: 1.5em;"">where the primary learning goal is not simply class label prediction. </span></p> <p><span style=""line-height: 1.5em;""></span><span style=""line-height: 1.5em;"">The proposed work will focus on two aims. Aim 1 is to construct novel machine learning methods </span><span style=""line-height: 1.5em;"">when the learning goal is more than mere class label prediction, such as class conditional probability </span><span style=""line-height: 1.5em;"">estimation. In Aim 2, I plan to tackle problems where the observations are not paired. For example, </span><span style=""line-height: 1.5em;"">in semi-supervised learning, only a small proportion of observations have their labels available. The </span><span style=""line-height: 1.5em;"">proposed study will have significant value in probability, machine learning and non-parametric statistics. </span><span style=""line-height: 1.5em;"">For HQP training, Master's and PhD students will be intensively involved in the research process. </span><span style=""line-height: 1.5em;"">They will gain solid experience in developing efficient algorithms, conducting numerical analysis, and </span><span style=""line-height: 1.5em;"">collaborating with scholars in many other disciplines.</span></p>