Canada Spends Logo

Explore our Budget 2026 analysis

Grant: Discovery Grants Program - Individual

McMaster University

$46,000.00

Variable Selection and Streaming Algorithms in Clustering Applications of Mixture Models

FY 2023-24

Summary
In plain language, clustering, also known as unsupervised classification, is the process of grouping observations into groups (or clusters) with similar characteristics. With the advent of the big data era, statistical approaches to clustering have proven superior to machine learning approaches in a variety of settings. This is in contrast to the supervised classification paradigm, where machine learning methods dominate. The past few years has seen significant leaps in statistical model-based approaches to clustering, led by several researchers worldwide including the applicant. Despite recent progress, some gaps remain in the suite of clustering methodologies available for modern datasets. The proposed research will address two major gaps. For one, explicit dimension reduction approaches -- i.e., variable selection approaches -- for data where some clusters may be skewed are underdeveloped. This has led to a situation whereby variable selection is all but ignored in many clustering analyses with real data. Another weakness is the dependence on the EM algorithm and, in particular, the absence of alternatives that are more efficient and better suited to large datasets. Typical EM algorithms, and batch algorithms in general, have access to the entire sample of data being analyzed, which is processed by the algorithm in a single batch. Streaming algorithms receive their data in a stream, one observation at a time and update the model parameters as the samples are received. Streaming algorithms will be developed for model-based clustering. Once established for multivariate data, both the variable selection and streaming work will be extended to the matrix-variate paradigm so that its impact may be felt in applications involving three-way data, e.g., in studies involving multivariate longitudinal data. As usual, efficient software will be made freely available so that the developed methodology can be used by others and applied to a wide variety of multivariate longitudinal data.
Program
Discovery Grants Program - Individual
Location
Ontario, Canada
Recipient Type
Academia
Type
Grant
Source
NSERC's Awards Data

As published

Application id
RGPIN-2023-06030
Name nom
McNicholas, Paul
Department département
Mathematics & Statistics
Organization id
27
Institution établissement
McMaster University
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2023
Competition year année de concours
2023
Award amount
46000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1508
Committee name en
Mathematics and Statistics
Committee name fr
Mathématiques et statistique
Area of application code
1101
Area of application group en
Health, education and social services
Area of application group fr
Santé publique, éducation et services sociaux
Area of application en
Human health (including medically-related psychological research)
Area of application fr
Santé publique (y compris recherche psychologique liée à la médecine)
Research subject code
3009
Research subject group en
Statistics and probability
Research subject group fr
Parametric inference
Research subject en
Statistique et probabilité
Research subject fr
Inférence paramétrique
Application title
Variable Selection and Streaming Algorithms in Clustering Applications of Mixture Models
Keyword
Clustering; Mixture models; Streaming algorithms; Variable selection
Application summary
In plain language, clustering, also known as unsupervised classification, is the process of grouping observations into groups (or clusters) with similar characteristics. With the advent of the big data era, statistical approaches to clustering have proven superior to machine learning approaches in a variety of settings. This is in contrast to the supervised classification paradigm, where machine learning methods dominate. The past few years has seen significant leaps in statistical model-based approaches to clustering, led by several researchers worldwide including the applicant. Despite recent progress, some gaps remain in the suite of clustering methodologies available for modern datasets. The proposed research will address two major gaps. For one, explicit dimension reduction approaches -- i.e., variable selection approaches -- for data where some clusters may be skewed are underdeveloped. This has led to a situation whereby variable selection is all but ignored in many clustering analyses with real data. Another weakness is the dependence on the EM algorithm and, in particular, the absence of alternatives that are more efficient and better suited to large datasets. Typical EM algorithms, and batch algorithms in general, have access to the entire sample of data being analyzed, which is processed by the algorithm in a single batch. Streaming algorithms receive their data in a stream, one observation at a time and update the model parameters as the samples are received. Streaming algorithms will be developed for model-based clustering. Once established for multivariate data, both the variable selection and streaming work will be extended to the matrix-variate paradigm so that its impact may be felt in applications involving three-way data, e.g., in studies involving multivariate longitudinal data. As usual, efficient software will be made freely available so that the developed methodology can be used by others and applied to a wide variety of multivariate longitudinal data.