Grant: Discovery Grants Program - Individual
York University
$27,000.00
Robust Nonparametric Methods for Complex Data
EF 2023-24
Résumé
Developing innovative methods to handle complex data, including data with outliers and/or privacy constraints, is the objective of the proposed research. In the modern age, institutions collect substantial amounts of complex data to enhance services, advance science and technology or increase efficiency. For, e.g., institutions track financial returns by the minute, monitor three-dimensional images of the brain by the second (f-MRI scans) and individuals' social networks over months. This data is imperfect, in the sense that it may contain mistakes such as mis-recorded values or abnormalities. These mistakes are called outliers, the presence of which can be hard to spot in such large datasets and be seriously damaging to the accuracy of statistical estimation. To build models that can handle outliers, we will first use a popular data analysis tool called depth measures. We will use depth measures in innovative ways, such as to detect sudden changes in data observed over time and to compare how treatments affect complex data outcomes in an experiment. In addition, complex data admit several kinds of outliers. For example, treatments may affect three-dimensional images of the brain in diverse ways. They may change the shape of the brain, the color or intensity in certain regions, or even the size of the brain. These diverse types of abnormalities affect statistical methods in diverse ways. We will use a combination of domain knowledge, classical statistics, and modern mathematics to uncover and categorize these diverse types of outliers. We will then develop modelling tools that can handle their presence. Complex data is further complicated by the necessity of protecting individuals' privacy. Researchers have uncovered a variety of attacks ill-meaning individuals can use to uncover information from datasets. They can execute these attacks even if so-called `identifying information,' such as names and addresses, have been removed. Differential privacy has become popular because it protects against these attacks. Differential privacy does this by obscuring data or model output. In this way, an adversary cannot tell if the output provides information about an individual they are targeting, or if the output is simply a result of obscuration. However, obscuration damages the accuracy of statistical estimation. Damage from obscuration is known as the cost of privacy. Major objectives of my proposal are quantifying the cost of privacy and creating methods which minimize the cost of privacy. To achieve such objectives, we will exploit the fact that handling outliers and enforcing differential privacy are similar problems. For example, one project is to repurpose data visualization tools that cannot be corrupted by outliers to develop graphical summaries of data which are differentially private. The development of differentially private methods allows these institutions to further society, without compromising the privacy of individuals.
Programme
Discovery Grants Program - Individual
Bénéficiaire
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data
Tel que publié
Application id
RGPIN-2023-05088
Name nom
Ramsay, Kelly
Department département
Mathematics and Statistics
Organization id
38
Institution établissement
York 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
27000
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
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
3010
Research subject group en
Statistics and probability
Research subject group fr
Nonparametric inference
Research subject en
Statistique et probabilité
Research subject fr
Inférence non paramétrique
Application title
Robust Nonparametric Methods for Complex Data
Keyword
change-point; cost of privacy; data depth; differential privacy; functional data; hypothesis testing; manifold-valued data; nonparametric statistics; privacy budget; robust statistics
Application summary
Developing innovative methods to handle complex data, including data with outliers and/or privacy constraints, is the objective of the proposed research. In the modern age, institutions collect substantial amounts of complex data to enhance services, advance science and technology or increase efficiency. For, e.g., institutions track financial returns by the minute, monitor three-dimensional images of the brain by the second (f-MRI scans) and individuals' social networks over months. This data is imperfect, in the sense that it may contain mistakes such as mis-recorded values or abnormalities. These mistakes are called outliers, the presence of which can be hard to spot in such large datasets and be seriously damaging to the accuracy of statistical estimation. To build models that can handle outliers, we will first use a popular data analysis tool called depth measures. We will use depth measures in innovative ways, such as to detect sudden changes in data observed over time and to compare how treatments affect complex data outcomes in an experiment. In addition, complex data admit several kinds of outliers. For example, treatments may affect three-dimensional images of the brain in diverse ways. They may change the shape of the brain, the color or intensity in certain regions, or even the size of the brain. These diverse types of abnormalities affect statistical methods in diverse ways. We will use a combination of domain knowledge, classical statistics, and modern mathematics to uncover and categorize these diverse types of outliers. We will then develop modelling tools that can handle their presence. Complex data is further complicated by the necessity of protecting individuals' privacy. Researchers have uncovered a variety of attacks ill-meaning individuals can use to uncover information from datasets. They can execute these attacks even if so-called `identifying information,' such as names and addresses, have been removed. Differential privacy has become popular because it protects against these attacks. Differential privacy does this by obscuring data or model output. In this way, an adversary cannot tell if the output provides information about an individual they are targeting, or if the output is simply a result of obscuration. However, obscuration damages the accuracy of statistical estimation. Damage from obscuration is known as the cost of privacy. Major objectives of my proposal are quantifying the cost of privacy and creating methods which minimize the cost of privacy. To achieve such objectives, we will exploit the fact that handling outliers and enforcing differential privacy are similar problems. For example, one project is to repurpose data visualization tools that cannot be corrupted by outliers to develop graphical summaries of data which are differentially private. The development of differentially private methods allows these institutions to further society, without compromising the privacy of individuals.